Playht delete circular buffer with mutex check (#123)

Signed-off-by: root <root@af6633a5cfe8>
Co-authored-by: root <root@af6633a5cfe8>

LICENSE_MIT

@@ -2,7 +2,7 @@ Under specific conditions that are described here: https://github.com/jambonz/fr
 
 The MIT License (MIT)
 
-Copyright (c) 2023 Drachtio Communications Services, LLC
+Copyright (c) 2024 FirstFive8, Inc
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

mod_assemblyai_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_audio_fork/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_audio_fork/Makefile.am

@@ -2,9 +2,17 @@ include $(top_srcdir)/build/modmake.rulesam
 MODNAME=mod_audio_fork
 
 mod_LTLIBRARIES = mod_audio_fork.la
-mod_audio_fork_la_SOURCES  = mod_audio_fork.c lws_glue.cpp parser.cpp audio_pipe.cpp
+mod_audio_fork_la_SOURCES  = mod_audio_fork.c lws_glue.cpp parser.cpp audio_pipe.cpp vector_math.cpp
 mod_audio_fork_la_CFLAGS   = $(AM_CFLAGS)
 mod_audio_fork_la_CXXFLAGS = $(AM_CXXFLAGS) -std=c++11
 
+if USE_AVX2
+mod_audio_fork_la_CXXFLAGS += -mavx2 -DUSE_AVX2
+else
+if USE_SSE2
+mod_audio_fork_la_CXXFLAGS += -msse2 -DUSE_SSE2
+endif
+endif
+
 mod_audio_fork_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
-mod_audio_fork_la_LDFLAGS  = -avoid-version -module -no-undefined -shared `pkg-config --libs libwebsockets` 
+mod_audio_fork_la_LDFLAGS  = -avoid-version -module -no-undefined -shared `pkg-config --libs libwebsockets` -lstdc++ -lboost_system -lboost_thread

mod_audio_fork/audio_pipe.cpp

@@ -1,4 +1,5 @@
 #include "audio_pipe.hpp"
+#include <switch.h>
 
 #include <cassert>
 #include <iostream>
@@ -64,13 +65,17 @@ int AudioPipe::lws_callback(struct lws *wsi,
           std::string username, password;
 
           ap->getBasicAuth(username, password);
-          lwsl_notice("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_APPEND_HANDSHAKE_HEADER username: %s, password: xxxxxx\n", username.c_str());
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_APPEND_HANDSHAKE_HEADER username: %s, password: xxxxxx\n", username.c_str());
           if (dch_lws_http_basic_auth_gen(username.c_str(), password.c_str(), b, sizeof(b))) break;
           if (lws_add_http_header_by_token(wsi, WSI_TOKEN_HTTP_AUTHORIZATION, (unsigned char *)b, strlen(b), p, end)) return -1;
         }
       }
       break;
 
+    case LWS_CALLBACK_WS_CLIENT_DROP_PROTOCOL:
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "AudioPipe::lws_service_thread LWS_CALLBACK_WS_CLIENT_DROP_PROTOCOL\n");
+      break;
+
     case LWS_CALLBACK_EVENT_WAIT_CANCELLED:
       processPendingConnects(vhd);
       processPendingDisconnects(vhd);
@@ -80,13 +85,13 @@ int AudioPipe::lws_callback(struct lws *wsi,
       {
         AudioPipe* ap = findAndRemovePendingConnect(wsi);
         int rc = lws_http_client_http_response(wsi);
-        lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR: %s, response status %d\n", in ? (char *)in : "(null)", rc); 
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR: %s, response status %d\n", in ? (char *)in : "(null)", rc); 
         if (ap) {
           ap->m_state = LWS_CLIENT_FAILED;
-          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_FAIL, (char *) in);
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_FAIL, (char *) in, NULL, len);
         }
         else {
-          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR unable to find wsi %p..\n", wsi); 
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR unable to find wsi %p..\n", wsi); 
         }
       }      
       break;
@@ -98,10 +103,10 @@ int AudioPipe::lws_callback(struct lws *wsi,
           *ppAp = ap;
           ap->m_vhd = vhd;
           ap->m_state = LWS_CLIENT_CONNECTED;
-          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_SUCCESS, NULL);
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_SUCCESS, NULL, NULL, len);
         }
         else {
-          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_ESTABLISHED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_ESTABLISHED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
         }
       }      
       break;
@@ -109,17 +114,17 @@ int AudioPipe::lws_callback(struct lws *wsi,
       {
         AudioPipe* ap = *ppAp;
         if (!ap) {
-          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CLOSED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CLOSED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
           return 0;
         }
         if (ap->m_state == LWS_CLIENT_DISCONNECTING) {
           // closed by us
-          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_CLOSED_GRACEFULLY, NULL);
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_CLOSED_GRACEFULLY, NULL, NULL, len);
         }
         else if (ap->m_state == LWS_CLIENT_CONNECTED) {
           // closed by far end
-          lwsl_notice("%s socket closed by far end\n", ap->m_uuid.c_str());
-          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_DROPPED, NULL);
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"%s socket closed by far end\n", ap->m_uuid.c_str());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_DROPPED, NULL, NULL, len);
         }
         ap->m_state = LWS_CLIENT_DISCONNECTED;
 
@@ -133,58 +138,71 @@ int AudioPipe::lws_callback(struct lws *wsi,
 
     case LWS_CALLBACK_CLIENT_RECEIVE:
       {
+        
         AudioPipe* ap = *ppAp;
         if (!ap) {
-          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          return 0;
+        }
+        
+        if (ap->m_state == LWS_CLIENT_DISCONNECTING) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"AudioPipe::lws_service_thread race condition: got incoming message while closing the connection.\n");
           return 0;
         }
 
         if (lws_frame_is_binary(wsi)) {
-          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE received binary frame, discarding.\n");
-          return 0;
-        }
-
-        if (lws_is_first_fragment(wsi)) {
-          // allocate a buffer for the entire chunk of memory needed
-          assert(nullptr == ap->m_recv_buf);
-          ap->m_recv_buf_len = len + lws_remaining_packet_payload(wsi);
-          ap->m_recv_buf = (uint8_t*) malloc(ap->m_recv_buf_len);
-          ap->m_recv_buf_ptr = ap->m_recv_buf;
-        }
-
-        size_t write_offset = ap->m_recv_buf_ptr - ap->m_recv_buf;
-        size_t remaining_space = ap->m_recv_buf_len - write_offset;
-        if (remaining_space < len) {
-          lwsl_notice("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE buffer realloc needed.\n");
-          size_t newlen = ap->m_recv_buf_len + RECV_BUF_REALLOC_SIZE;
-          if (newlen > MAX_RECV_BUF_SIZE) {
-            free(ap->m_recv_buf);
-            ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
-            ap->m_recv_buf_len = 0;
-            lwsl_notice("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE max buffer exceeded, truncating message.\n");
+          if (len > 0 && ap->is_bidirectional_audio_stream()) {
+            ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::BINARY, NULL, (char *) in, len);
+          } else if (len > 0) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE received unexpected binary frame, discarding.\n");
           }
           else {
-            ap->m_recv_buf = (uint8_t*) realloc(ap->m_recv_buf, newlen);
-            if (nullptr != ap->m_recv_buf) {
-              ap->m_recv_buf_len = newlen;
-              ap->m_recv_buf_ptr = ap->m_recv_buf + write_offset;
-            }
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE received zero length binary frame, discarding.\n");
           }
         }
-
-        if (nullptr != ap->m_recv_buf) {
-          if (len > 0) {
-            memcpy(ap->m_recv_buf_ptr, in, len);
-            ap->m_recv_buf_ptr += len;
+        else {
+          if (lws_is_first_fragment(wsi)) {
+            // allocate a buffer for the entire chunk of memory needed
+            assert(nullptr == ap->m_recv_buf);
+            ap->m_recv_buf_len = len + lws_remaining_packet_payload(wsi);
+            ap->m_recv_buf = (uint8_t*) malloc(ap->m_recv_buf_len);
+            ap->m_recv_buf_ptr = ap->m_recv_buf;
           }
-          if (lws_is_final_fragment(wsi)) {
-            if (nullptr != ap->m_recv_buf) {
-              std::string msg((char *)ap->m_recv_buf, ap->m_recv_buf_ptr - ap->m_recv_buf);
-              ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::MESSAGE, msg.c_str());
-              if (nullptr != ap->m_recv_buf) free(ap->m_recv_buf);
+
+          size_t write_offset = ap->m_recv_buf_ptr - ap->m_recv_buf;
+          size_t remaining_space = ap->m_recv_buf_len - write_offset;
+          if (remaining_space < len) {
+            //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE buffer realloc needed.\n");
+            size_t newlen = ap->m_recv_buf_len + RECV_BUF_REALLOC_SIZE;
+            if (newlen > MAX_RECV_BUF_SIZE) {
+              free(ap->m_recv_buf);
+              ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
+              ap->m_recv_buf_len = 0;
+              switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE max buffer exceeded, truncating message.\n");
+            }
+            else {
+              ap->m_recv_buf = (uint8_t*) realloc(ap->m_recv_buf, newlen);
+              if (nullptr != ap->m_recv_buf) {
+                ap->m_recv_buf_len = newlen;
+                ap->m_recv_buf_ptr = ap->m_recv_buf + write_offset;
+              }
+            }
+          }
+
+          if (nullptr != ap->m_recv_buf) {
+            if (len > 0) {
+              memcpy(ap->m_recv_buf_ptr, in, len);
+              ap->m_recv_buf_ptr += len;
+            }
+            if (lws_is_final_fragment(wsi)) {
+              if (nullptr != ap->m_recv_buf) {
+                std::string msg((char *)ap->m_recv_buf, ap->m_recv_buf_ptr - ap->m_recv_buf);
+                ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::MESSAGE, msg.c_str(), NULL, len);
+                if (nullptr != ap->m_recv_buf) free(ap->m_recv_buf);
+              }
+              ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
+              ap->m_recv_buf_len = 0;
             }
-            ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
-            ap->m_recv_buf_len = 0;
           }
         }
       }
@@ -194,13 +212,13 @@ int AudioPipe::lws_callback(struct lws *wsi,
       {
         AudioPipe* ap = *ppAp;
         if (!ap) {
-          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_WRITEABLE %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_WRITEABLE %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
           return 0;
         }
 
         // check for graceful close - send a zero length binary frame
         if (ap->isGracefulShutdown()) {
-          lwsl_notice("%s graceful shutdown - sending zero length binary frame to flush any final responses\n", ap->m_uuid.c_str());
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"%s graceful shutdown - sending zero length binary frame to flush any final responses\n", ap->m_uuid.c_str());
           std::lock_guard<std::mutex> lk(ap->m_audio_mutex);
           int sent = lws_write(wsi, (unsigned char *) ap->m_audio_buffer + LWS_PRE, 0, LWS_WRITE_BINARY);
           return 0;
@@ -209,20 +227,21 @@ int AudioPipe::lws_callback(struct lws *wsi,
         // check for text frames to send
         {
           std::lock_guard<std::mutex> lk(ap->m_text_mutex);
-          if (ap->m_metadata.length() > 0) {
-            uint8_t buf[ap->m_metadata.length() + LWS_PRE];
-            memcpy(buf + LWS_PRE, ap->m_metadata.c_str(), ap->m_metadata.length());
-            int n = ap->m_metadata.length();
+          if (!ap->m_metadata_list.empty()) {
+            const std::string& message = ap->m_metadata_list.front();
+            uint8_t buf[message.length() + LWS_PRE];
+            memcpy(buf + LWS_PRE, message.c_str(), message.length());
+            int n = message.length();
             int m = lws_write(wsi, buf + LWS_PRE, n, LWS_WRITE_TEXT);
-            ap->m_metadata.clear();
+
             if (m < n) {
-              return -1;
+              return -1; // Failed to send the full message
             }
 
-            // there may be audio data, but only one write per writeable event
-            // get it next time
+            // Remove the message that was successfully sent
+            ap->m_metadata_list.pop_front();
+            // Request another writable event if there are more messages
             lws_callback_on_writable(wsi);
-
             return 0;
           }
         }
@@ -239,7 +258,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
             size_t datalen = ap->m_audio_buffer_write_offset - LWS_PRE;
             int sent = lws_write(wsi, (unsigned char *) ap->m_audio_buffer + LWS_PRE, datalen, LWS_WRITE_BINARY);
             if (sent < datalen) {
-              lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_WRITEABLE %s attemped to send %lu only sent %d wsi %p..\n", 
+              switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_WRITEABLE %s attemped to send %lu only sent %d wsi %p..\n", 
                 ap->m_uuid.c_str(), datalen, sent, wsi); 
             }
             ap->m_audio_buffer_write_offset = LWS_PRE;
@@ -368,7 +387,7 @@ void AudioPipe::addPendingConnect(AudioPipe* ap) {
   {
     std::lock_guard<std::mutex> guard(mutex_connects);
     pendingConnects.push_back(ap);
-    lwsl_notice("%s after adding connect there are %lu pending connects\n", 
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG,"%s after adding connect there are %lu pending connects\n", 
       ap->m_uuid.c_str(), pendingConnects.size());
   }
   lws_cancel_service(context);
@@ -378,7 +397,7 @@ void AudioPipe::addPendingDisconnect(AudioPipe* ap) {
   {
     std::lock_guard<std::mutex> guard(mutex_disconnects);
     pendingDisconnects.push_back(ap);
-    lwsl_notice("%s after adding disconnect there are %lu pending disconnects\n", 
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG,"%s after adding disconnect there are %lu pending disconnects\n", 
       ap->m_uuid.c_str(), pendingDisconnects.size());
   }
   lws_cancel_service(ap->m_vhd->context);
@@ -417,11 +436,11 @@ bool AudioPipe::lws_service_thread() {
   info.timeout_secs_ah_idle = 10;       // secs to allow a client to hold an ah without using it
   info.retry_and_idle_policy = &retry;
 
-  lwsl_notice("AudioPipe::lws_service_thread creating context\n");
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"AudioPipe::lws_service_thread creating context\n");
 
   context = lws_create_context(&info);
   if (!context) {
-    lwsl_err("AudioPipe::lws_service_thread failed creating context\n"); 
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread failed creating context\n"); 
     return false;
   }
 
@@ -438,16 +457,16 @@ bool AudioPipe::lws_service_thread() {
 
 void AudioPipe::initialize(const char* protocol, int loglevel, log_emit_function logger) {
   protocolName = protocol;
-  //lws_set_log_level(loglevel, logger);
+  lws_set_log_level(loglevel, logger);
 
-  lwsl_notice("AudioPipe::initialize starting\n"); 
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE,"AudioPipe::initialize starting\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = false;
   serviceThread = std::thread(&AudioPipe::lws_service_thread);
 }
 
 bool AudioPipe::deinitialize() {
-  lwsl_notice("AudioPipe::deinitialize\n"); 
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE,"AudioPipe::deinitialize\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = true;
   if (serviceThread.joinable()) {
@@ -458,7 +477,8 @@ bool AudioPipe::deinitialize() {
 
 // instance members
 AudioPipe::AudioPipe(const char* uuid, const char* host, unsigned int port, const char* path,
-  int sslFlags, size_t bufLen, size_t minFreespace, const char* username, const char* password, char* bugname, notifyHandler_t callback) :
+  int sslFlags, size_t bufLen, size_t minFreespace, const char* username, const char* password, char* bugname,
+  int bidirectional_audio_stream, notifyHandler_t callback) :
   m_uuid(uuid), m_host(host), m_port(port), m_path(path), m_sslFlags(sslFlags),
   m_audio_buffer_min_freespace(minFreespace), m_audio_buffer_max_len(bufLen), m_gracefulShutdown(false),
   m_audio_buffer_write_offset(LWS_PRE), m_recv_buf(nullptr), m_recv_buf_ptr(nullptr), m_bugname(bugname),
@@ -468,12 +488,12 @@ AudioPipe::AudioPipe(const char* uuid, const char* host, unsigned int port, cons
     m_username.assign(username);
     m_password.assign(password);
   }
-
+  m_bidirectional_audio_stream = bidirectional_audio_stream;
   m_audio_buffer = new uint8_t[m_audio_buffer_max_len];
 }
 AudioPipe::~AudioPipe() {
   if (m_audio_buffer) delete [] m_audio_buffer;
-  if (m_recv_buf) delete [] m_recv_buf;
+  if (m_recv_buf) free(m_recv_buf);
 }
 
 void AudioPipe::connect(void) {
@@ -501,7 +521,7 @@ bool AudioPipe::connect_client(struct lws_per_vhost_data *vhd) {
   m_vhd = vhd;
 
   m_wsi = lws_client_connect_via_info(&i);
-  lwsl_notice("%s attempting connection, wsi is %p\n", m_uuid.c_str(), m_wsi);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG,"%s attempting connection, wsi is %p\n", m_uuid.c_str(), m_wsi);
 
   return nullptr != m_wsi;
 }
@@ -510,7 +530,7 @@ void AudioPipe::bufferForSending(const char* text) {
   if (m_state != LWS_CLIENT_CONNECTED) return;
   {
     std::lock_guard<std::mutex> lk(m_text_mutex);
-    m_metadata.append(text);
+    m_metadata_list.emplace_back(text);
   }
   addPendingWrite(this);
 }

mod_audio_fork/audio_pipe.hpp

@@ -27,10 +27,11 @@ namespace drachtio {
       CONNECT_FAIL,
       CONNECTION_DROPPED,
       CONNECTION_CLOSED_GRACEFULLY,
-      MESSAGE
+      MESSAGE,
+      BINARY
     };
     typedef void (*log_emit_function)(int level, const char *line);
-    typedef void (*notifyHandler_t)(const char *sessionId, const char* bugname, NotifyEvent_t event, const char* message);
+    typedef void (*notifyHandler_t)(const char *sessionId, const char* bugname, NotifyEvent_t event, const char* message, const char* binary, size_t binary_len );
 
     struct lws_per_vhost_data {
       struct lws_context *context;
@@ -44,7 +45,8 @@ namespace drachtio {
 
     // constructor
     AudioPipe(const char* uuid, const char* host, unsigned int port, const char* path, int sslFlags, 
-      size_t bufLen, size_t minFreespace, const char* username, const char* password, char* bugname, notifyHandler_t callback);
+      size_t bufLen, size_t minFreespace, const char* username, const char* password, char* bugname,
+      int bidirectional_audio, notifyHandler_t callback);
     ~AudioPipe();  
 
     LwsState_t getLwsState(void) { return m_state; }
@@ -83,6 +85,10 @@ namespace drachtio {
       return m_gracefulShutdown;
     }
 
+    bool is_bidirectional_audio_stream() {
+      return m_bidirectional_audio_stream;
+    }
+
     void close() ;
 
     // no default constructor or copying
@@ -124,7 +130,7 @@ namespace drachtio {
     std::string m_bugname;
     unsigned int m_port;
     std::string m_path;
-    std::string m_metadata;
+    std::list<std::string> m_metadata_list;
     std::mutex m_text_mutex;
     std::mutex m_audio_mutex;
     int m_sslFlags;
@@ -142,6 +148,7 @@ namespace drachtio {
     std::string m_username;
     std::string m_password;
     bool m_gracefulShutdown;
+    bool m_bidirectional_audio_stream;
   };
 
 } // namespace drachtio

mod_audio_fork/lws_glue.cpp

@@ -17,9 +17,18 @@
 #include "parser.hpp"
 #include "mod_audio_fork.h"
 #include "audio_pipe.hpp"
+#include "vector_math.h"
+
+#include <boost/circular_buffer.hpp>
+
+
+typedef boost::circular_buffer<uint16_t> CircularBuffer_t;
 
 #define RTP_PACKETIZATION_PERIOD 20
 #define FRAME_SIZE_8000  320 /*which means each 20ms frame as 320 bytes at 8 khz (1 channel only)*/
+#define BUFFER_GROW_SIZE (16384)
+#define AUDIO_MARKER 0xFFFF
+#define MAX_MARKS (30)
 
 namespace {
   static const char *requestedBufferSecs = std::getenv("MOD_AUDIO_FORK_BUFFER_SECS");
@@ -31,14 +40,179 @@ namespace {
   static unsigned int idxCallCount = 0;
   static uint32_t playCount = 0;
 
+  static bool markCountExceeded(private_t* tech_pvt) {
+    if (nullptr != tech_pvt->pVecMarksInUse) {
+      std::deque<std::string>* pVecMarksInUse = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInUse);
+      std::deque<std::string>* pVecMarksInInventory = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+      return pVecMarksInUse->size()+ pVecMarksInInventory->size() >= MAX_MARKS;
+    }
+    return false;
+  }
+
+  switch_status_t processIncomingBinary(private_t* tech_pvt, switch_core_session_t* session, const char* message, size_t dataLength) {
+    std::vector<uint8_t> data;
+
+    // Prepend the set-aside byte if there is one
+    if (tech_pvt->has_set_aside_byte) {
+        data.push_back(tech_pvt->set_aside_byte);
+        tech_pvt->has_set_aside_byte = false;
+    }
+
+    // Append the new incoming message
+    data.insert(data.end(), message, message + dataLength);
+
+    // Check if the total data length is now odd
+    if (data.size() % 2 != 0) {
+        // Set aside the last byte
+        tech_pvt->set_aside_byte = data.back();
+        tech_pvt->has_set_aside_byte = true;
+        data.pop_back(); // Remove the last byte from the data vector
+    }
+
+    // Convert the data to 16-bit elements
+    const uint16_t* data_uint16 = reinterpret_cast<const uint16_t*>(data.data());
+    size_t numSamples = data.size() / sizeof(uint16_t);
+
+    // Access the prebuffer
+    CircularBuffer_t* cBuffer = static_cast<CircularBuffer_t*>(tech_pvt->streamingPreBuffer);
+
+    int numMarkers = 0;
+    std::deque<std::string>* pVecMarksInInventory = nullptr;
+    std::deque<std::string>* pVecMarksInUse = nullptr;
+    if (nullptr != tech_pvt->pVecMarksInInventory) {
+      pVecMarksInInventory = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+      pVecMarksInUse = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInUse);
+      numMarkers = pVecMarksInInventory->size();
+
+      // move inventory to in-use
+      pVecMarksInUse->insert(pVecMarksInUse->end(), pVecMarksInInventory->begin(), pVecMarksInInventory->end());
+      pVecMarksInInventory->clear();
+    }
+  
+    // Ensure the prebuffer has enough capacity
+    if (cBuffer->capacity() - cBuffer->size() < numSamples + numMarkers) {
+        size_t newCapacity = cBuffer->size() + std::max(numSamples + numMarkers, (size_t)BUFFER_GROW_SIZE);
+        cBuffer->set_capacity(newCapacity);
+    }
+
+    // prepend any markers
+    while (numMarkers-- > 0) {
+      cBuffer->push_back(AUDIO_MARKER);
+    }
+
+    // Append the data to the prebuffer
+    cBuffer->insert(cBuffer->end(), data_uint16, data_uint16 + numSamples);
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Appended %zu 16-bit samples to the prebuffer.\n", numSamples);
+
+    // if we haven't reached threshold amount of prebuffered data, return
+    if (cBuffer->size() < tech_pvt->streamingPreBufSize) {
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Prebuffered data is below threshold %u, returning.\n", tech_pvt->streamingPreBufSize);
+        return SWITCH_STATUS_SUCCESS;
+    }
+
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Prebuffered data samples %u is above threshold %u, prepare to playout.\n", cBuffer->size(), tech_pvt->streamingPreBufSize);
+
+    // after initial pre-buffering, rachet down the threshold to 40ms
+    tech_pvt->streamingPreBufSize = 320 * tech_pvt->downscale_factor * 2;
+
+    // Check for downsampling factor
+    size_t downsample_factor = tech_pvt->downscale_factor;
+
+    // Calculate the number of samples that can be evenly divided by the downsample factor
+    size_t numCompleteSamples = (cBuffer->size() / downsample_factor) * downsample_factor;
+
+    // Handle leftover samples
+    std::vector<uint16_t> leftoverSamples;
+    size_t numLeftoverSamples = cBuffer->size() - numCompleteSamples;
+    if (numLeftoverSamples > 0) {
+        leftoverSamples.assign(cBuffer->end() - numLeftoverSamples, cBuffer->end());
+        cBuffer->resize(numCompleteSamples);
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Temporarily removing %u leftover samples due to downsampling.\n", numLeftoverSamples);
+    }
+
+    // resample if necessary
+    std::vector<int16_t> out;
+    try {
+      if (tech_pvt->bidirectional_audio_resampler) {
+        // Improvement: Use assign to convert circular buffer to vector for resampling
+        std::vector<int16_t> in;
+        in.assign(cBuffer->begin(), cBuffer->end()); 
+        out.resize(in.size() * 6); // max upsampling would be from 8k -> 48k
+
+        spx_uint32_t in_len = in.size();
+        spx_uint32_t out_len = out.size();
+
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Resampling %u samples into a buffer that can hold %u samples\n", in.size(), out_len);
+
+        speex_resampler_process_interleaved_int(tech_pvt->bidirectional_audio_resampler, in.data(), &in_len, out.data(), &out_len);
+
+        // Resize the output buffer to match the output length from resampler
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Resizing output buffer from %u to %u samples\n", in.size(), out_len);
+
+        out.resize(out_len);
+      }
+      else {
+        // If no resampling is needed, copy the data from the prebuffer to the output buffer
+        out.assign(cBuffer->begin(), cBuffer->end());
+      }
+    } catch (const std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error resampling incoming binary message: %s\n", e.what());
+      return SWITCH_STATUS_FALSE;
+    } catch (...) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error resampling incoming binary message\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    if (nullptr != tech_pvt->mutex && switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
+      CircularBuffer_t *playoutBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
+
+      try {
+        // Resize the buffer if necessary
+        if (playoutBuffer->capacity() - playoutBuffer->size() < out.size()) {
+          size_t newCapacity = playoutBuffer->size() + std::max(out.size(), (size_t)BUFFER_GROW_SIZE);
+          playoutBuffer->set_capacity(newCapacity);
+          //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Resized playout buffer to new capacity: %zu\n", newCapacity);
+        }
+        // Push the data into the buffer.
+        playoutBuffer->insert(playoutBuffer->end(), out.begin(), out.end());
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Appended %zu 16-bit samples to the playout buffer.\n", out.size());
+      } catch (const std::exception& e) {
+        switch_mutex_unlock(tech_pvt->mutex);
+        cBuffer->clear();
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error processing incoming binary message: %s\n", e.what());
+        return SWITCH_STATUS_FALSE;
+      } catch (...) {
+        switch_mutex_unlock(tech_pvt->mutex);
+        cBuffer->clear();
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error processing incoming binary message\n");
+        return SWITCH_STATUS_FALSE;
+      }
+      switch_mutex_unlock(tech_pvt->mutex);
+      cBuffer->clear();
+
+      // Put the leftover samples back in the prebuffer for the next time
+      if (!leftoverSamples.empty()) {
+          cBuffer->insert(cBuffer->end(), leftoverSamples.begin(), leftoverSamples.end());
+          //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Put back %u leftover samples into the prebuffer.\n", leftoverSamples.size());
+      }
+      return SWITCH_STATUS_SUCCESS;
+    }
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Failed to get mutext (temp)\n");
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
   void processIncomingMessage(private_t* tech_pvt, switch_core_session_t* session, const char* message) {
     std::string msg = message;
     std::string type;
     cJSON* json = parse_json(session, msg, type) ;
     if (json) {
-      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) processIncomingMessage - received %s message\n", tech_pvt->id, type.c_str());
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) processIncomingMessage - received %s message %s\n", tech_pvt->id, type.c_str(), message);
       cJSON* jsonData = cJSON_GetObjectItem(json, "data");
-      if (0 == type.compare("playAudio")) {
+      if (0 == type.compare("playAudio") &&
+        // playAudio is enabled and there is no bidirectional audio from stream is enabled.
+        tech_pvt->bidirectional_audio_enable &&
+        !tech_pvt->bidirectional_audio_stream) {
         if (jsonData) {
           // dont send actual audio bytes in event message
           cJSON* jsonFile = NULL;
@@ -120,6 +294,66 @@ namespace {
         // kill any current playback on the channel
         switch_channel_t *channel = switch_core_session_get_channel(session);
         switch_channel_set_flag_value(channel, CF_BREAK, 2);
+
+        // this will dump buffered incoming audio
+        tech_pvt->clear_bidirectional_audio_buffer = true;
+      }
+      else if (0 == type.compare("mark")) {
+        cJSON* data = cJSON_GetObjectItem(json, "data");
+        if (data) {
+          cJSON* name = cJSON_GetObjectItem(data, "name");
+          if (cJSON_IsString(name) && name->valuestring) {
+            if (markCountExceeded(tech_pvt)) {
+              switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "(%u) processIncomingMessage - mark count exceeded, discarding mark %s\n", tech_pvt->id, cJSON_GetStringValue(name));
+            }
+            else {
+              if (nullptr == tech_pvt->pVecMarksInInventory) {
+                tech_pvt->pVecMarksInInventory = static_cast<void *>(new std::deque<std::string>());
+                tech_pvt->pVecMarksInUse = static_cast<void *>(new std::deque<std::string>());
+                tech_pvt->pVecMarksCleared = static_cast<void *>(new std::deque<std::string>());
+              }
+              std::deque<std::string>* pVec = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+              pVec->push_back(name->valuestring);
+            }
+          }
+        }
+      }
+      else if (0 == type.compare("clearMarks")) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) processIncomingMessage - received clearMarks\n", tech_pvt->id);
+        if (nullptr != tech_pvt->pVecMarksInInventory) {
+          std::deque<std::string>* pVecMarksInInventory = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+          std::deque<std::string>* pVecMarksInUse = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInUse);
+          std::deque<std::string>* pVecMarksCleared = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksCleared);
+          pVecMarksCleared->insert(pVecMarksCleared->end(), pVecMarksInUse->begin(), pVecMarksInUse->end());
+          pVecMarksCleared->insert(pVecMarksCleared->end(), pVecMarksInInventory->begin(), pVecMarksInInventory->end());
+          pVecMarksInInventory->clear();
+          pVecMarksInUse->clear();
+        }
+      }
+      else if (0 == type.compare("transcription")) {
+        char* jsonString = cJSON_PrintUnformatted(jsonData);
+        tech_pvt->responseHandler(session, EVENT_TRANSCRIPTION, jsonString);
+        free(jsonString);        
+      }
+      else if (0 == type.compare("transfer")) {
+        char* jsonString = cJSON_PrintUnformatted(jsonData);
+        tech_pvt->responseHandler(session, EVENT_TRANSFER, jsonString);
+        free(jsonString);                
+      }
+      else if (0 == type.compare("disconnect")) {
+        char* jsonString = cJSON_PrintUnformatted(jsonData);
+        tech_pvt->responseHandler(session, EVENT_DISCONNECT, jsonString);
+        free(jsonString);        
+      }
+      else if (0 == type.compare("error")) {
+        char* jsonString = cJSON_PrintUnformatted(jsonData);
+        tech_pvt->responseHandler(session, EVENT_ERROR, jsonString);
+        free(jsonString);        
+      }
+      else if (0 == type.compare("json")) {
+        char* jsonString = cJSON_PrintUnformatted(json);
+        tech_pvt->responseHandler(session, EVENT_JSON, jsonString);
+        free(jsonString);
       }
       else if (0 == type.compare("transcription")) {
         char* jsonString = cJSON_PrintUnformatted(jsonData);
@@ -156,7 +390,7 @@ namespace {
     }
   }
 
-  static void eventCallback(const char* sessionId, const char* bugname, drachtio::AudioPipe::NotifyEvent_t event, const char* message) {
+  static void eventCallback(const char* sessionId, const char* bugname, drachtio::AudioPipe::NotifyEvent_t event, const char* message, const char* binary, size_t len) {
     switch_core_session_t* session = switch_core_session_locate(sessionId);
     if (session) {
       switch_channel_t *channel = switch_core_session_get_channel(session);
@@ -198,6 +432,9 @@ namespace {
             case drachtio::AudioPipe::MESSAGE:
               processIncomingMessage(tech_pvt, session, message);
             break;
+            case drachtio::AudioPipe::BINARY:
+            processIncomingBinary(tech_pvt, session, binary, len);
+            break;
           }
         }
       }
@@ -206,11 +443,13 @@ namespace {
   }
   switch_status_t fork_data_init(private_t *tech_pvt, switch_core_session_t *session, char * host, 
     unsigned int port, char* path, int sslFlags, int sampling, int desiredSampling, int channels, 
-    char *bugname, char* metadata, responseHandler_t responseHandler) {
+    char *bugname, char* metadata, int bidirectional_audio_enable,
+    int bidirectional_audio_stream, int bidirectional_audio_sample_rate, responseHandler_t responseHandler) {
 
     const char* username = nullptr;
     const char* password = nullptr;
     int err;
+    int bidirectional_audio_stream_enable = bidirectional_audio_enable + bidirectional_audio_stream;
     switch_codec_implementation_t read_impl;
     switch_channel_t *channel = switch_core_session_get_channel(session);
 
@@ -234,13 +473,30 @@ namespace {
     tech_pvt->buffer_overrun_notified = 0;
     tech_pvt->audio_paused = 0;
     tech_pvt->graceful_shutdown = 0;
+    tech_pvt->streamingPlayoutBuffer = (void *) new CircularBuffer_t(8192);
+    tech_pvt->bidirectional_audio_enable = bidirectional_audio_enable;
+    tech_pvt->bidirectional_audio_stream = bidirectional_audio_stream;
+    tech_pvt->bidirectional_audio_sample_rate = bidirectional_audio_sample_rate;
+    tech_pvt->clear_bidirectional_audio_buffer = false;
+    tech_pvt->has_set_aside_byte = 0;
+    tech_pvt->downscale_factor = 1;
+    if (bidirectional_audio_sample_rate > sampling) {
+      tech_pvt->downscale_factor = bidirectional_audio_sample_rate / sampling;
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "downscale_factor is %d\n", tech_pvt->downscale_factor);
+    }
+    tech_pvt->streamingPreBufSize = 320 * tech_pvt->downscale_factor * 4; // min 80ms prebuffer
+    tech_pvt->streamingPreBuffer = (void *) new CircularBuffer_t(8192);
+    tech_pvt->pVecMarksInInventory = nullptr;
+    tech_pvt->pVecMarksInUse = nullptr;
+    tech_pvt->pVecMarksCleared = nullptr;
+
     strncpy(tech_pvt->bugname, bugname, MAX_BUG_LEN);
     if (metadata) strncpy(tech_pvt->initialMetadata, metadata, MAX_METADATA_LEN);
     
     size_t buflen = LWS_PRE + (FRAME_SIZE_8000 * desiredSampling / 8000 * channels * 1000 / RTP_PACKETIZATION_PERIOD * nAudioBufferSecs);
 
     drachtio::AudioPipe* ap = new drachtio::AudioPipe(tech_pvt->sessionId, host, port, path, sslFlags, 
-      buflen, read_impl.decoded_bytes_per_packet, username, password, bugname, eventCallback);
+      buflen, read_impl.decoded_bytes_per_packet, username, password, bugname, bidirectional_audio_stream_enable, eventCallback);
     if (!ap) {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error allocating AudioPipe\n");
       return SWITCH_STATUS_FALSE;
@@ -262,6 +518,15 @@ namespace {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) no resampling needed for this call\n", tech_pvt->id);
     }
 
+    if (bidirectional_audio_sample_rate && sampling != bidirectional_audio_sample_rate) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) bidirectional audio resampling from %u to %u, channels %d\n", tech_pvt->id, bidirectional_audio_sample_rate, sampling, channels);
+      tech_pvt->bidirectional_audio_resampler = speex_resampler_init(1, bidirectional_audio_sample_rate, sampling, SWITCH_RESAMPLE_QUALITY, &err);
+      if (0 != err) {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing bidirectional audio resampler: %s.\n", speex_resampler_strerror(err));
+        return SWITCH_STATUS_FALSE;
+      }
+    }
+
     switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) fork_data_init\n", tech_pvt->id);
 
     return SWITCH_STATUS_SUCCESS;
@@ -273,10 +538,47 @@ namespace {
       speex_resampler_destroy(tech_pvt->resampler);
       tech_pvt->resampler = nullptr;
     }
+    if (tech_pvt->bidirectional_audio_resampler) {
+      speex_resampler_destroy(tech_pvt->bidirectional_audio_resampler);
+      tech_pvt->bidirectional_audio_resampler = nullptr;
+    }
     if (tech_pvt->mutex) {
       switch_mutex_destroy(tech_pvt->mutex);
       tech_pvt->mutex = nullptr;
     }
+    if (tech_pvt->streamingPlayoutBuffer) {
+      CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
+      delete cBuffer;
+      tech_pvt->streamingPlayoutBuffer = nullptr;
+    }
+    if (tech_pvt->streamingPreBuffer) {
+      CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPreBuffer;
+      delete cBuffer;
+      tech_pvt->streamingPreBuffer = nullptr;
+    }
+
+    if (tech_pvt->pVecMarksInInventory) {
+      delete static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+      tech_pvt->pVecMarksInInventory = nullptr;
+      delete static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInUse);
+      tech_pvt->pVecMarksInUse = nullptr;
+      delete static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksCleared);
+      tech_pvt->pVecMarksCleared = nullptr;
+    }
+  }
+
+  static void send_mark_event(private_t* tech_pvt, const char* name, int cleared = false) {
+    drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe *>(tech_pvt->pAudioPipe);
+    std::ostringstream json;
+    json << "{\"type\": \"mark\", \"data\": {\"name\":\"" << name << "\", ";
+    if (cleared) json << "\"event\": \"cleared\"}}";
+    else json << "\"event\": \"playout\"}}";
+
+    if (pAudioPipe) {
+      std::string str = json.str();
+      pAudioPipe->bufferForSending(str.c_str());
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) send_mark_event: %s\n", tech_pvt->id, str.c_str());
+    }
   }
 
   void lws_logger(int level, const char *line) {
@@ -341,7 +643,10 @@ extern "C" {
     }
 
     std::string strHost(server + offset);
-    std::regex re("^(.+?):?(\\d+)?(/.*)?$");
+    //- `([^/:]+)` captures the hostname/IP address, match any character except in the set
+    //- `:?([0-9]*)?` optionally captures a colon and the port number, if it's present.
+    //- `(/.*)` captures everything else (the path).
+    std::regex re("([^/:]+):?([0-9]*)?(/.*)?$");
     std::smatch matches;
     if(std::regex_search(strHost, matches, re)) {
       /*
@@ -372,8 +677,8 @@ extern "C" {
     switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_audio_fork: audio buffer (in secs):    %d secs\n", nAudioBufferSecs);
     switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_audio_fork: sub-protocol:              %s\n", mySubProtocolName);
  
-    int logs = LLL_ERR | LLL_WARN | LLL_NOTICE ;
-     //LLL_INFO | LLL_PARSER | LLL_HEADER | LLL_EXT | LLL_CLIENT  | LLL_LATENCY | LLL_DEBUG ;
+    //int logs = LLL_ERR | LLL_WARN | LLL_NOTICE | LLL_INFO | LLL_PARSER | LLL_HEADER | LLL_EXT | LLL_CLIENT  | LLL_LATENCY | LLL_DEBUG ;
+    int logs = LLL_ERR | LLL_WARN | LLL_NOTICE;
     drachtio::AudioPipe::initialize(mySubProtocolName, logs, lws_logger);
     switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_audio_fork successfully initialized\n");
     return SWITCH_STATUS_SUCCESS;
@@ -392,18 +697,22 @@ extern "C" {
   }
 
   switch_status_t fork_session_init(switch_core_session_t *session, 
-              responseHandler_t responseHandler,
-              uint32_t samples_per_second, 
-              char *host,
-              unsigned int port,
-              char *path,
-              int sampling,
-              int sslFlags,
-              int channels,
-              char *bugname,
-              char* metadata, 
-              void **ppUserData)
-  {    	
+    responseHandler_t responseHandler,
+    uint32_t samples_per_second, 
+    char *host,
+    unsigned int port,
+    char *path,
+    int sampling,
+    int sslFlags,
+    int channels,
+    char *bugname,
+    char* metadata,
+    int bidirectional_audio_enable,
+    int bidirectional_audio_stream,
+    int bidirectional_audio_sample_rate,
+    void **ppUserData
+    )
+  {
     int err;
 
     // allocate per-session data structure
@@ -412,8 +721,9 @@ extern "C" {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "error allocating memory!\n");
       return SWITCH_STATUS_FALSE;
     }
+
     if (SWITCH_STATUS_SUCCESS != fork_data_init(tech_pvt, session, host, port, path, sslFlags, samples_per_second, sampling, channels, 
-      bugname, metadata, responseHandler)) {
+      bugname, metadata, bidirectional_audio_enable, bidirectional_audio_stream, bidirectional_audio_sample_rate, responseHandler)) {
       destroy_tech_pvt(tech_pvt);
       return SWITCH_STATUS_FALSE;
     }
@@ -619,5 +929,147 @@ extern "C" {
     return SWITCH_TRUE;
   }
 
+  switch_bool_t dub_speech_frame(switch_media_bug_t *bug, private_t* tech_pvt) {
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
+    if (switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
+
+      // if flag was set to clear the buffer, do so and clear the flag
+      if (tech_pvt->clear_bidirectional_audio_buffer) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) dub_speech_frame - clearing buffer\n", tech_pvt->id); 
+        cBuffer->clear();
+        tech_pvt->clear_bidirectional_audio_buffer = false;
+
+        // send "mark" event for any queued markers
+        if (nullptr != tech_pvt->pVecMarksInInventory) {
+          std::deque<std::string>* pVecMarksInInventory = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+          std::deque<std::string>* pVecMarksInUse = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInUse);
+          if (pVecMarksInInventory->size() + pVecMarksInUse->size() > 0) {
+            std::deque<std::string> vec = *pVecMarksInUse;
+            vec.insert(vec.end(), pVecMarksInInventory->begin(), pVecMarksInInventory->end());
+            for (auto it = vec.begin(); it != vec.end(); ++it) {
+              switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "(%u) dub_speech_frame - Marker %s cleared\n",
+                  tech_pvt->id, it->c_str());
+              send_mark_event(tech_pvt, it->c_str(), true);
+            }
+
+            // put the "in-use" ones into the "cleared" queue so we dont notify again when they eventually come through
+            std::deque<std::string>* pVecMarksCleared = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksCleared);
+            pVecMarksCleared->insert(pVecMarksCleared->end(), pVecMarksInUse->begin(), pVecMarksInUse->end());
+
+            pVecMarksInUse->clear();
+            pVecMarksInInventory->clear();
+          }
+        }
+      }
+      else {
+        switch_frame_t* rframe = switch_core_media_bug_get_write_replace_frame(bug);
+        int16_t *fp = reinterpret_cast<int16_t*>(rframe->data);
+
+        rframe->channels = 1;
+        rframe->datalen = rframe->samples * sizeof(int16_t);
+
+        int16_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
+        memset(data, 0, sizeof(data));
+
+        int samplesToCopy = std::min(static_cast<int>(cBuffer->size()), static_cast<int>(rframe->samples));
+
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) dub_speech_frame - samples to copy %u\n", tech_pvt->id, samplesToCopy); 
+
+        bool hasMarkers = false;
+        std::deque<std::string>* pVecInventory = nullptr;
+        std::deque<std::string>* pVecInUse = nullptr;
+        std::deque<std::string>* pVecCleared = nullptr;
+        if (tech_pvt->pVecMarksInInventory && tech_pvt->pVecMarksInUse && tech_pvt->pVecMarksCleared) {
+          pVecInventory = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInInventory);
+          pVecInUse = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksInUse);
+          pVecCleared = static_cast<std::deque<std::string>*>(tech_pvt->pVecMarksCleared);
+          hasMarkers = pVecInUse->size() + pVecCleared->size() >  0;
+        }
+
+        if (hasMarkers) {
+          /* discard markers and send notifications */
+          auto bufferIter = cBuffer->begin();
+          auto dataIter = data;
+          for (int i = 0; i < samplesToCopy; ++i) {
+            if (*bufferIter == AUDIO_MARKER) {
+              // Marker detected, discard it and send a notice unless it was previously cleared
+              auto * pVec = pVecCleared->size() > 0 ? pVecCleared : pVecInUse;
+              if (!pVec->empty()) {
+                auto name = pVec->front();
+                pVec->pop_front();
+
+                if (pVec == pVecInUse) {
+                  send_mark_event(tech_pvt, name.c_str());
+                  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) dub_speech_frame - Marker %s detected in playout\n",
+                    tech_pvt->id, name.c_str());
+                }
+                else {
+                  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) dub_speech_frame - Marker %s detected in playout but previously cleared\n",
+                    tech_pvt->id, name.c_str());
+                }
+              }
+            } else {
+              // Copy valid audio samplewhat 
+              *dataIter = *bufferIter;
+              ++dataIter;
+            }
+            ++bufferIter;
+          }
+
+          // Remove the processed samples (including discarded markers) from the buffer
+          cBuffer->erase(cBuffer->begin(), cBuffer->begin() + samplesToCopy);
+
+          // Adjust the number of samples copied to the output frame
+          int validSamplesCopied = std::distance(data, dataIter);
+
+          if (validSamplesCopied > 0) {
+            vector_add(fp, data, validSamplesCopied);
+          }
+        }
+        else {
+          std::copy_n(cBuffer->begin(), samplesToCopy, data);
+          cBuffer->erase(cBuffer->begin(), cBuffer->begin() + samplesToCopy);
+
+          if (samplesToCopy > 0) {
+            vector_add(fp, data, rframe->samples);
+          } else if (pVecInventory && pVecInventory->size()) {
+            // no bidirectional audio to dub but still have some mark in inventory, send them now
+            auto name = pVecInventory->front();
+            pVecInventory->pop_front();
+
+            send_mark_event(tech_pvt, name.c_str());
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "(%u) dub_speech_frame - Marker %s detected in inventory\n",
+              tech_pvt->id, name.c_str());
+          }
+        }
+        vector_normalize(fp, rframe->samples);
+
+        switch_core_media_bug_set_write_replace_frame(bug, rframe);
+      }
+      switch_mutex_unlock(tech_pvt->mutex);
+    }
+    return SWITCH_TRUE;
+  }
+
+  switch_status_t fork_session_stop_play(switch_core_session_t *session, char *bugname) {
+    switch_channel_t *channel = switch_core_session_get_channel(session);
+    switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, bugname);
+    if (!bug) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "fork_session_stop_play failed because no bug\n");
+      return SWITCH_STATUS_FALSE;
+    }
+    private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
+
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
+
+    if (switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
+      if (cBuffer != nullptr) {
+        cBuffer->clear();
+      }
+      switch_mutex_unlock(tech_pvt->mutex);
+    }
+    return SWITCH_STATUS_SUCCESS;
+  }
+
 }
 

mod_audio_fork/lws_glue.h

@@ -8,13 +8,16 @@ int parse_ws_uri(switch_channel_t *channel, const char* szServerUri, char* host,
 switch_status_t fork_init();
 switch_status_t fork_cleanup();
 switch_status_t fork_session_init(switch_core_session_t *session, responseHandler_t responseHandler,
-		uint32_t samples_per_second, char *host, unsigned int port, char* path, int sampling, int sslFlags, int channels, 
-    char *bugname, char* metadata, void **ppUserData);
+  uint32_t samples_per_second, char *host, unsigned int port, char* path, int sampling, int sslFlags, int channels, 
+  char *bugname, char* metadata, int bidirectional_audio_enable,
+  int bidirectional_audio_stream, int bidirectional_audio_sample_rate, void **ppUserData);
 switch_status_t fork_session_cleanup(switch_core_session_t *session, char *bugname, char* text, int channelIsClosing);
+switch_status_t fork_session_stop_play(switch_core_session_t *session, char *bugname);
 switch_status_t fork_session_pauseresume(switch_core_session_t *session, char *bugname, int pause);
 switch_status_t fork_session_graceful_shutdown(switch_core_session_t *session, char *bugname);
 switch_status_t fork_session_send_text(switch_core_session_t *session, char *bugname, char* text);
 switch_bool_t fork_frame(switch_core_session_t *session, switch_media_bug_t *bug);
+switch_bool_t dub_speech_frame(switch_media_bug_t *bug, private_t * tech_pvt);
 switch_status_t fork_service_threads();
 switch_status_t fork_session_connect(void **ppUserData);
 #endif

mod_audio_fork/mod_audio_fork.c

@@ -27,40 +27,48 @@ static void responseHandler(switch_core_session_t* session, const char * eventNa
 
 static switch_bool_t capture_callback(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type)
 {
+	switch_bool_t ret = SWITCH_TRUE;
 	switch_core_session_t *session = switch_core_media_bug_get_session(bug);
+	private_t* tech_pvt = (private_t *)  switch_core_media_bug_get_user_data(bug);
 	switch (type) {
 	case SWITCH_ABC_TYPE_INIT:
 		break;
 
 	case SWITCH_ABC_TYPE_CLOSE:
 		{
-      private_t* tech_pvt = (private_t *)  switch_core_media_bug_get_user_data(bug);
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "Got SWITCH_ABC_TYPE_CLOSE for bug %s\n", tech_pvt->bugname);
       fork_session_cleanup(session, tech_pvt->bugname, NULL, 1);
 		}
 		break;
 	
 	case SWITCH_ABC_TYPE_READ:
-		return fork_frame(session, bug);
+		ret = fork_frame(session, bug);
 		break;
 
+	case SWITCH_ABC_TYPE_WRITE_REPLACE:
+		ret = dub_speech_frame(bug, tech_pvt);
+	break;
+
 	case SWITCH_ABC_TYPE_WRITE:
 	default:
 		break;
 	}
 
-	return SWITCH_TRUE;
+	return ret;
 }
 
 static switch_status_t start_capture(switch_core_session_t *session, 
-        switch_media_bug_flag_t flags, 
-        char* host,
-        unsigned int port, 
-        char* path,
-        int sampling,
-        int sslFlags,
-	      char* bugname, 
-        char* metadata)
+	switch_media_bug_flag_t flags, 
+	char* host,
+	unsigned int port, 
+	char* path,
+	int sampling,
+	int sslFlags,
+	int bidirectional_audio_enable,
+	int bidirectional_audio_stream,
+	int bidirectional_audio_sample_rate,
+	char* bugname, 
+	char* metadata)
 {
 	switch_channel_t *channel = switch_core_session_get_channel(session);
 	switch_media_bug_t *bug;
@@ -71,8 +79,8 @@ static switch_status_t start_capture(switch_core_session_t *session,
   int channels = (flags & SMBF_STEREO) ? 2 : 1;
 
 	switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, 
-    "mod_audio_fork (%s): streaming %d sampling to %s path %s port %d tls: %s.\n", 
-    bugname, sampling, host, path, port, sslFlags ? "yes" : "no");
+    "mod_audio_fork (%s): streaming %d sampling to %s path %s port %d tls: %s bidirectional_audio_sample_rate: %d.\n", 
+    bugname, sampling, host, path, port, sslFlags ? "yes" : "no", bidirectional_audio_sample_rate);
 
 	if (switch_channel_get_private(channel, bugname)) {
 		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "mod_audio_fork: bug %s already attached!\n", bugname);
@@ -88,10 +96,12 @@ static switch_status_t start_capture(switch_core_session_t *session,
 
 	switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "calling fork_session_init.\n");
 	if (SWITCH_STATUS_FALSE == fork_session_init(session, responseHandler, read_codec->implementation->actual_samples_per_second, 
-		host, port, path, sampling, sslFlags, channels, bugname, metadata, &pUserData)) {
+		host, port, path, sampling, sslFlags, channels, bugname, metadata, bidirectional_audio_enable, bidirectional_audio_stream,
+		bidirectional_audio_sample_rate, &pUserData)) {
 		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing mod_audio_fork session.\n");
 		return SWITCH_STATUS_FALSE;
 	}
+
 	switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "adding bug %s.\n", bugname);
 	if ((status = switch_core_media_bug_add(session, bugname, NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
 		return status;
@@ -133,6 +143,16 @@ static switch_status_t do_pauseresume(switch_core_session_t *session, char* bugn
 	return status;
 }
 
+static switch_status_t stop_play(switch_core_session_t *session, char* bugname)
+{
+	switch_status_t status = SWITCH_STATUS_SUCCESS;
+
+	switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "mod_audio_fork stop_play\n");
+	status = fork_session_stop_play(session, bugname);
+
+	return status;
+}
+
 static switch_status_t do_graceful_shutdown(switch_core_session_t *session, char* bugname)
 {
 	switch_status_t status = SWITCH_STATUS_SUCCESS;
@@ -159,10 +179,10 @@ static switch_status_t send_text(switch_core_session_t *session, char* bugname,
   return status;
 }
 
-#define FORK_API_SYNTAX "<uuid> [start | stop | send_text | pause | resume | graceful-shutdown ] [wss-url | path] [mono | mixed | stereo] [8000 | 16000 | 24000 | 32000 | 64000] [bugname] [metadata]"
+#define FORK_API_SYNTAX "<uuid> [start | stop | send_text | pause | resume | graceful-shutdown | stop_play ] [wss-url | path] [mono | mixed | stereo] [8000 | 16000 | 24000 | 32000 | 64000] [bugname] [metadata] [bidirectionalAudio_enabled] [bidirectionalAudio_stream_enabled] [bidirectionalAudio_stream_samplerate]"
 SWITCH_STANDARD_API(fork_function)
 {
-	char *mycmd = NULL, *argv[7] = { 0 };
+	char *mycmd = NULL, *argv[10] = { 0 };
 	int argc = 0;
 	switch_status_t status = SWITCH_STATUS_FALSE;
   char *bugname = MY_BUG_NAME;
@@ -196,6 +216,9 @@ SWITCH_STANDARD_API(fork_function)
         }
 				status = do_stop(lsession, bugname, text);
       }
+			else if (!strcasecmp(argv[1], "stop_play")) {
+				status = stop_play(lsession, bugname);
+			}
 			else if (!strcasecmp(argv[1], "pause")) {
         if (argc > 2) bugname = argv[2];
 				status = do_pauseresume(lsession, bugname, 1);
@@ -230,10 +253,31 @@ SWITCH_STANDARD_API(fork_function)
         char host[MAX_WS_URL_LEN], path[MAX_PATH_LEN];
         unsigned int port;
         int sslFlags;
+				
         int sampling = 8000;
-      	switch_media_bug_flag_t flags = SMBF_READ_STREAM ;
+      	switch_media_bug_flag_t flags = SMBF_READ_STREAM;
         char *metadata = NULL;
-        if( argc > 6) {
+				int bidirectional_audio_enable = 1;
+				int bidirectional_audio_stream = 0;
+				int bidirectional_audio_sample_rate = 0;
+				// Expecting that bidirectional audio params is always received together with bugname and metadata even they are empty string
+				if (argc > 9) {
+					if (argv[5][0] != '\0') {
+						bugname = argv[5];
+					}
+					if (argv[6][0] != '\0') {
+						metadata = argv[6];
+					}
+					bidirectional_audio_enable = !strcmp(argv[7], "true") ? 1 : 0;
+					bidirectional_audio_stream = !strcmp(argv[8], "true") ? 1 : 0;
+					bidirectional_audio_sample_rate = atoi(argv[9]);
+
+					if (bidirectional_audio_enable &&
+						bidirectional_audio_stream &&
+						bidirectional_audio_sample_rate) {
+						flags |= SMBF_WRITE_REPLACE ;
+					}
+				} else if( argc > 6 ) {
           bugname = argv[5];
           metadata = argv[6];
         }
@@ -241,6 +285,7 @@ SWITCH_STANDARD_API(fork_function)
           if (argv[5][0] == '{' || argv[5][0] == '[') metadata = argv[5];
           else bugname = argv[5];
         }
+
         if (0 == strcmp(argv[3], "mixed")) {
           flags |= SMBF_WRITE_STREAM ;
         }
@@ -268,7 +313,8 @@ SWITCH_STANDARD_API(fork_function)
 				else if (sampling % 8000 != 0) {
           switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "invalid sample rate: %s\n", argv[4]);					
 				}
-        status = start_capture(lsession, flags, host, port, path, sampling, sslFlags, bugname, metadata);
+        status = start_capture(lsession, flags, host, port, path, sampling, sslFlags,
+					bidirectional_audio_enable, bidirectional_audio_stream, bidirectional_audio_sample_rate, bugname, metadata);
 			}
       else {
         switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "unsupported mod_audio_fork cmd: %s\n", argv[1]);

mod_audio_fork/mod_audio_fork.h

@@ -52,6 +52,24 @@ struct private_data {
   int audio_paused:1;
   int graceful_shutdown:1;
   char initialMetadata[8192];
+
+  // for "mark" feature of bidirectional audio
+  void *pVecMarksInInventory;
+  void *pVecMarksInUse;
+  void *pVecMarksCleared;
+
+  // bidirectional audio
+  void *streamingPlayoutBuffer;
+  void *streamingPreBuffer;
+  int streamingPreBufSize;
+  uint8_t set_aside_byte;
+  int has_set_aside_byte;
+  int downscale_factor;
+  SpeexResamplerState *bidirectional_audio_resampler;
+  int bidirectional_audio_enable;
+	int bidirectional_audio_stream;
+  int bidirectional_audio_sample_rate;
+  int clear_bidirectional_audio_buffer;
 };
 
 typedef struct private_data private_t;

mod_audio_fork/vector_math.cpp

@@ -0,0 +1,213 @@
+#include "vector_math.h"
+#include <assert.h>
+#include <string.h>
+#include <cstdlib>
+
+#define GRANULAR_VOLUME_MAX (50)
+#define SMAX 32767
+#define SMIN (-32768)
+#define normalize_to_16bit_basic(n) if (n > SMAX) n = SMAX; else if (n < SMIN) n = SMIN;
+#define normalize_volume_granular(x) if (x > GRANULAR_VOLUME_MAX) x = GRANULAR_VOLUME_MAX; if (x < -GRANULAR_VOLUME_MAX) x = -GRANULAR_VOLUME_MAX;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#if defined(USE_AVX2)
+#include <immintrin.h>
+#pragma message("Using AVX2 SIMD.")
+void vector_add(int16_t* a, int16_t* b, size_t len) {
+    size_t i = 0;
+    for (; i + 15 < len; i += 16) {
+        __m256i va = _mm256_loadu_si256((const __m256i*)(a + i));
+        __m256i vb = _mm256_loadu_si256((const __m256i*)(b + i));
+        __m256i vc = _mm256_add_epi16(va, vb);
+        _mm256_storeu_si256((__m256i*)(a + i), vc);
+    }
+    for (; i < len; ++i) {
+        a[i] += b[i];
+    }
+}
+void vector_normalize(int16_t* a, size_t len) {
+  __m256i max_val = _mm256_set1_epi16(SMAX);
+  __m256i min_val = _mm256_set1_epi16(SMIN);
+
+  size_t i = 0;
+  for (; i + 15 < len; i += 16) {
+      __m256i values = _mm256_loadu_si256((__m256i*)(a + i));
+      __m256i gt_max = _mm256_cmpgt_epi16(values, max_val);
+      __m256i lt_min = _mm256_cmpgt_epi16(min_val, values);
+      values = _mm256_blendv_epi8(values, max_val, gt_max);
+      values = _mm256_blendv_epi8(values, min_val, lt_min);
+      _mm256_storeu_si256((__m256i*)(a + i), values);
+  }
+
+  // Process remaining elements
+  for (; i < len; ++i) {
+      if (a[i] > SMAX) a[i] = SMAX;
+      else if (a[i] < SMIN) a[i] = SMIN;
+  }
+}
+typedef union {
+    int16_t* data;
+    __m256i* fp_avx2;
+} vector_data_t;
+
+void vector_change_sln_volume_granular(int16_t* data, uint32_t samples, int32_t vol) {
+  float newrate = 0;
+  static const float pos[GRANULAR_VOLUME_MAX] = {
+		1.122018,   1.258925,   1.412538,   1.584893,   1.778279,   1.995262,   2.238721,   2.511887,   2.818383,   3.162278,
+		3.548134,   3.981072,   4.466835,   5.011872,   5.623413,   6.309574,   7.079458,   7.943282,   8.912509,  10.000000,
+		11.220183,  12.589254,  14.125375,  15.848933,  17.782795,  19.952621,  22.387213,  25.118862,  28.183832,  31.622776,
+		35.481335,  39.810719,  44.668358,  50.118729,  56.234131,  63.095726,  70.794586,  79.432816,  89.125107, 100.000000,
+		112.201836, 125.892517, 141.253784, 158.489334, 177.827942, 199.526215, 223.872070, 251.188705, 281.838318, 316.227753
+  };
+  static const float neg[GRANULAR_VOLUME_MAX] = {
+		0.891251, 0.794328, 0.707946, 0.630957, 0.562341, 0.501187, 0.446684, 0.398107, 0.354813, 0.316228,
+		0.281838, 0.251189, 0.223872, 0.199526, 0.177828, 0.158489, 0.141254, 0.125893, 0.112202, 0.100000,
+		0.089125, 0.079433, 0.070795, 0.063096, 0.056234, 0.050119, 0.044668, 0.039811, 0.035481, 0.031623,
+		0.028184, 0.025119, 0.022387, 0.019953, 0.017783, 0.015849, 0.014125, 0.012589, 0.011220, 0.010000,
+		0.008913, 0.007943, 0.007079, 0.006310, 0.005623, 0.005012, 0.004467, 0.003981, 0.003548, 0.000000  // NOTE mapped -50 dB ratio to total silence instead of 0.003162
+  };
+  const float* chart;
+  uint32_t i = abs(vol) - 1;
+
+  if (vol == 0) return;
+  normalize_volume_granular(vol);
+
+  chart = vol > 0 ? pos : neg;
+
+  newrate = chart[i];
+
+  if (newrate) {
+    __m256 scale_factor_reg = _mm256_set1_ps(newrate);
+    uint32_t processed_samples = samples - (samples % 8); // Ensure we process only multiples of 8
+    for (uint32_t i = 0; i < processed_samples; i += 8) {
+      __m128i data_ = _mm_loadu_si128((__m128i*)(data + i));
+      __m256i data_32 = _mm256_cvtepi16_epi32(data_);
+
+      __m256 data_float = _mm256_cvtepi32_ps(data_32);
+      __m256 result = _mm256_mul_ps(data_float, scale_factor_reg);
+
+      __m256i result_32 = _mm256_cvtps_epi32(result);
+
+      // Handle saturation
+      __m256i min_val = _mm256_set1_epi32(SMIN);
+      __m256i max_val = _mm256_set1_epi32(SMAX);
+      result_32 = _mm256_min_epi32(result_32, max_val);
+      result_32 = _mm256_max_epi32(result_32, min_val);
+
+      __m128i result_16 = _mm_packs_epi32(_mm256_castsi256_si128(result_32), _mm256_extractf128_si256(result_32, 1));
+
+      _mm_storeu_si128((__m128i*)(data + i), result_16);
+    }
+
+    // Process any remaining samples
+    for (uint32_t i = processed_samples; i < samples; i++) {
+      int32_t tmp = (int32_t)(data[i] * newrate);
+      tmp = tmp > SMAX ? SMAX : (tmp < SMIN ? SMIN : tmp);
+      data[i] = (int16_t)tmp;
+    }
+  }
+}
+#elif defined(USE_SSE2)
+#include <emmintrin.h>
+#pragma message("Using SSE2 SIMD.")
+void vector_add(int16_t* a, int16_t* b, size_t len) {
+  size_t i = 0;
+  for (; i + 7 < len; i += 8) {
+    __m128i va = _mm_loadu_si128((const __m128i*)(a + i));
+    __m128i vb = _mm_loadu_si128((const __m128i*)(b + i));
+    __m128i vc = _mm_add_epi16(va, vb);
+    _mm_storeu_si128((__m128i*)(a + i), vc);
+  }
+  for (; i < len; ++i) {
+    a[i] += b[i];
+  }
+}
+void vector_normalize(int16_t* a, size_t len) {
+  __m128i max_val = _mm_set1_epi16(SMAX);
+  __m128i min_val = _mm_set1_epi16(SMIN);
+
+  size_t i = 0;
+  for (; i + 7 < len; i += 8) {
+    __m128i values = _mm_loadu_si128((__m128i*)(a + i));
+    __m128i gt_max = _mm_cmpgt_epi16(values, max_val);
+    __m128i lt_min = _mm_cmpgt_epi16(min_val, values);
+    __m128i max_masked = _mm_and_si128(gt_max, max_val);
+    __m128i min_masked = _mm_and_si128(lt_min, min_val);
+    __m128i other_masked = _mm_andnot_si128(_mm_or_si128(gt_max, lt_min), values);
+    values = _mm_or_si128(_mm_or_si128(max_masked, min_masked), other_masked);
+    _mm_storeu_si128((__m128i*)(a + i), values);
+  }
+
+  // Process remaining elements
+  for (; i < len; ++i) {
+    if (a[i] > SMAX) a[i] = SMAX;
+    else if (a[i] < SMIN) a[i] = SMIN;
+  }
+}
+
+typedef union {
+    int16_t* data;
+    __m128i* fp_sse2;
+} vector_data_t;
+
+#else
+#pragma message("Building without vector math support")
+void vector_add(int16_t* a, int16_t* b, size_t len) {
+    for (size_t i = 0; i < len; i++) {
+        a[i] += b[i];
+    }
+}
+void vector_normalize(int16_t* a, size_t len) {
+  for (size_t i = 0; i < len; i++) {
+    normalize_to_16bit_basic(a[i]);
+  }
+}
+void vector_change_sln_volume_granular(int16_t* data, uint32_t samples, int32_t vol) {
+  float newrate = 0;
+  static const float pos[GRANULAR_VOLUME_MAX] = {
+		1.122018,   1.258925,   1.412538,   1.584893,   1.778279,   1.995262,   2.238721,   2.511887,   2.818383,   3.162278,
+		3.548134,   3.981072,   4.466835,   5.011872,   5.623413,   6.309574,   7.079458,   7.943282,   8.912509,  10.000000,
+		11.220183,  12.589254,  14.125375,  15.848933,  17.782795,  19.952621,  22.387213,  25.118862,  28.183832,  31.622776,
+		35.481335,  39.810719,  44.668358,  50.118729,  56.234131,  63.095726,  70.794586,  79.432816,  89.125107, 100.000000,
+		112.201836, 125.892517, 141.253784, 158.489334, 177.827942, 199.526215, 223.872070, 251.188705, 281.838318, 316.227753
+  };
+  static const float neg[GRANULAR_VOLUME_MAX] = {
+		0.891251, 0.794328, 0.707946, 0.630957, 0.562341, 0.501187, 0.446684, 0.398107, 0.354813, 0.316228,
+		0.281838, 0.251189, 0.223872, 0.199526, 0.177828, 0.158489, 0.141254, 0.125893, 0.112202, 0.100000,
+		0.089125, 0.079433, 0.070795, 0.063096, 0.056234, 0.050119, 0.044668, 0.039811, 0.035481, 0.031623,
+		0.028184, 0.025119, 0.022387, 0.019953, 0.017783, 0.015849, 0.014125, 0.012589, 0.011220, 0.010000,
+		0.008913, 0.007943, 0.007079, 0.006310, 0.005623, 0.005012, 0.004467, 0.003981, 0.003548, 0.000000  // NOTE mapped -50 dB ratio to total silence instead of 0.003162
+  };
+  const float* chart;
+  uint32_t i;
+
+  if (vol == 0) return;
+
+  normalize_volume_granular(vol);
+
+  chart = vol > 0 ? pos : neg;
+
+  i = abs(vol) - 1;
+  assert(i < GRANULAR_VOLUME_MAX);
+  newrate = chart[i];
+  if (newrate) {
+    int32_t tmp;
+    uint32_t x;
+    int16_t *fp = data;
+
+    for (x = 0; x < samples; x++) {
+      tmp = (int32_t) (fp[x] * newrate);
+      normalize_to_16bit_basic(tmp);
+      fp[x] = (int16_t) tmp;
+    }
+  }
+}
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif

mod_audio_fork/vector_math.h

@@ -0,0 +1,20 @@
+#ifndef VECTOR_MATH_H
+#define VECTOR_MATH_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void vector_add(int16_t* a, int16_t* b, size_t len);
+void vector_normalize(int16_t* a, size_t len);
+void vector_change_sln_volume_granular(int16_t* data, uint32_t samples, int32_t vol);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

mod_aws_lex/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_aws_lex/Makefile.am

@@ -7,4 +7,4 @@ mod_aws_lex_la_CFLAGS   = $(AM_CFLAGS)
 mod_aws_lex_la_CXXFLAGS = $(AM_CXXFLAGS) -std=c++11 -I${switch_srcdir}/libs/aws-sdk-cpp/aws-cpp-sdk-core/include -I${switch_srcdir}/libs/aws-sdk-cpp/aws-cpp-sdk-lexv2-runtime/include -I${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/include
 
 mod_aws_lex_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
-mod_aws_lex_la_LDFLAGS  = -avoid-version -module -no-undefined -L${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/lib -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-core -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-lexv2-runtime -laws-cpp-sdk-lexv2-runtime -laws-cpp-sdk-core -laws-c-event-stream -laws-checksums -laws-c-common -lpthread -lcurl -lcrypto -lssl -lz
+mod_aws_lex_la_LDFLAGS  = -avoid-version -module -no-undefined -L/usr/local/lib -L${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/lib -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-core -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-lexv2-runtime -laws-cpp-sdk-lexv2-runtime -laws-cpp-sdk-core -laws-c-event-stream -laws-checksums -laws-c-common -lpthread -lcurl -lcrypto -lssl -lz

mod_aws_lex/aws_lex_glue.cpp

@@ -173,13 +173,12 @@ public:
 		char *metadata,
 		const char* awsAccessKeyId, 
 		const char* awsSecretAccessKey,
+		const char* awsSessionToken,
 		responseHandler_t responseHandler,
 		errorHandler_t  errorHandler) : 
 	m_bot(bot), m_alias(alias), m_region(region), m_sessionId(sessionId), m_finished(false), m_finishing(false), m_packets(0),
 	m_pStream(nullptr), m_bPlayDone(false), m_bDiscardAudio(false)
 	{
-		Aws::String key(awsAccessKeyId);
-		Aws::String secret(awsSecretAccessKey);
 		Aws::String awsLocale(locale);
 		Aws::Client::ClientConfiguration config;
 		config.region = region;
@@ -190,8 +189,11 @@ public:
 		for (int i = 4; i < 20; i++) keySnippet[i] = 'x';
 		keySnippet[19] = '\0';
 
-		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p ACCESS_KEY_ID %s\n", this, keySnippet);		
-		if (*awsAccessKeyId && *awsSecretAccessKey) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p ACCESS_KEY_ID %s\n", this, keySnippet);
+		if (*awsAccessKeyId && *awsSecretAccessKey && *awsSessionToken) {
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "using AWS creds %s %s %s\n", awsAccessKeyId, awsSecretAccessKey, awsSessionToken);	
+			m_client = Aws::MakeUnique<LexRuntimeV2Client>(ALLOC_TAG, AWSCredentials(awsAccessKeyId, awsSecretAccessKey, awsSessionToken), config);
+		} else if (*awsAccessKeyId && *awsSecretAccessKey) {
 			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "using AWS creds %s %s\n", awsAccessKeyId, awsSecretAccessKey);	
 			m_client = Aws::MakeUnique<LexRuntimeV2Client>(ALLOC_TAG, AWSCredentials(awsAccessKeyId, awsSecretAccessKey), config);
 		}
@@ -540,7 +542,7 @@ static void *SWITCH_THREAD_FUNC lex_thread(switch_thread_t *thread, void *obj) {
 	bool ok = true;
 	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "lex_thread: starting cb %p\n", (void *) cb);
 	GStreamer* pStreamer = new GStreamer(cb->sessionId, cb->bot, cb->alias, cb->region, cb->locale, 
-		cb->intent, cb->metadata, cb->awsAccessKeyId, cb->awsSecretAccessKey, 
+		cb->intent, cb->metadata, cb->awsAccessKeyId, cb->awsSecretAccessKey, cb->awsSessionToken,
 		cb->responseHandler, cb->errorHandler);
 	if (!pStreamer) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "lex_thread: Error allocating streamer\n");
@@ -641,6 +643,7 @@ extern "C" {
 		memset(cb, sizeof(cb), 0);
 		const char* awsAccessKeyId = switch_channel_get_variable(channel, "AWS_ACCESS_KEY_ID");
 		const char* awsSecretAccessKey = switch_channel_get_variable(channel, "AWS_SECRET_ACCESS_KEY");
+		const char* awsSessionToken = switch_channel_get_variable(channel, "AWS_SESSION_TOKEN");
 
 		if (!hasDefaultCredentials && (!awsAccessKeyId || !awsSecretAccessKey)) {
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, 
@@ -654,6 +657,7 @@ extern "C" {
 		if (awsAccessKeyId && awsSecretAccessKey) {
 			strncpy(cb->awsAccessKeyId, awsAccessKeyId, 128);
 			strncpy(cb->awsSecretAccessKey, awsSecretAccessKey, 128);
+			if (awsSessionToken) strncpy(cb->awsSessionToken, awsSessionToken, 1024);
 		}
 		else {
 			strncpy(cb->awsAccessKeyId, std::getenv("AWS_ACCESS_KEY_ID"), 128);

mod_aws_lex/mod_aws_lex.h

@@ -30,6 +30,7 @@ struct cap_cb {
 	char sessionId[256];
   char awsAccessKeyId[128];
   char awsSecretAccessKey[128];
+	char awsSessionToken[1024];
   SpeexResamplerState *resampler;
 	void* streamer;
 	responseHandler_t responseHandler;

mod_aws_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_aws_transcribe/Makefile.am

@@ -7,4 +7,4 @@ mod_aws_transcribe_la_CFLAGS   = $(AM_CFLAGS)
 mod_aws_transcribe_la_CXXFLAGS = $(AM_CXXFLAGS) -std=c++11 -I${switch_srcdir}/libs/aws-sdk-cpp/aws-cpp-sdk-core/include -I${switch_srcdir}/libs/aws-sdk-cpp/aws-cpp-sdk-transcribestreaming/include -I${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/include
 
 mod_aws_transcribe_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
-mod_aws_transcribe_la_LDFLAGS  = -avoid-version -module -no-undefined -L${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/lib -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-core -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-transcribestreaming -laws-cpp-sdk-transcribestreaming -laws-cpp-sdk-core -laws-c-event-stream -laws-checksums -laws-c-common -lpthread -lcurl -lcrypto -lssl -lz
+mod_aws_transcribe_la_LDFLAGS  = -avoid-version -module -no-undefined -L/usr/local/lib -L${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/lib -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-core -L${switch_srcdir}/libs/aws-sdk-cpp/build/aws-cpp-sdk-transcribestreaming -laws-cpp-sdk-transcribestreaming -laws-cpp-sdk-core -laws-c-event-stream -laws-checksums -laws-c-common -laws-crt-cpp -lpthread -lcurl -lcrypto -lssl -lz

mod_aws_transcribe/aws_transcribe_glue.cpp

@@ -43,18 +43,17 @@ public:
     const char *sessionId,
 		const char *bugname,
 		u_int16_t channels,
-    char *lang, 
+    char *lang,
     int interim,
 		uint32_t samples_per_second,
-		const char* region, 
-		const char* awsAccessKeyId, 
+		const char* region,
+		const char* awsAccessKeyId,
 		const char* awsSecretAccessKey,
+		const char* awsSessionToken,
 		responseHandler_t responseHandler
   ) : m_sessionId(sessionId), m_bugname(bugname), m_finished(false), m_interim(interim), m_finishing(false), m_connected(false), m_connecting(false),
-	 		m_packets(0), m_responseHandler(responseHandler), m_pStream(nullptr), 
+	 		m_packets(0), m_responseHandler(responseHandler), m_pStream(nullptr),
 			m_audioBuffer(320 * (samples_per_second == 8000 ? 1 : 2), 15) {
-		Aws::String key(awsAccessKeyId);
-		Aws::String secret(awsSecretAccessKey);
 		Aws::Client::ClientConfiguration config;
 		if (region != nullptr && strlen(region) > 0) config.region = region;
 		char keySnippet[20];
@@ -63,14 +62,16 @@ public:
 		for (int i = 4; i < 20; i++) keySnippet[i] = 'x';
 		keySnippet[19] = '\0';
 
-		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p ACCESS_KEY_ID %s, region %s\n", this, keySnippet, region);		
-		if (*awsAccessKeyId && *awsSecretAccessKey) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p ACCESS_KEY_ID %s, region %s\n", this, keySnippet, region);
+		if (*awsAccessKeyId && *awsSecretAccessKey && *awsSessionToken) {
+			m_client = Aws::MakeUnique<TranscribeStreamingServiceClient>(ALLOC_TAG, AWSCredentials(awsAccessKeyId, awsSecretAccessKey, awsSessionToken), config);
+		} else if (*awsAccessKeyId && *awsSecretAccessKey) {
 			m_client = Aws::MakeUnique<TranscribeStreamingServiceClient>(ALLOC_TAG, AWSCredentials(awsAccessKeyId, awsSecretAccessKey), config);
 		}
 		else {
 			m_client = Aws::MakeUnique<TranscribeStreamingServiceClient>(ALLOC_TAG, config);
 		}
-	
+
     m_handler.SetTranscriptEventCallback([this](const TranscriptEvent& ev)
     {
 			switch_core_session_t* psession = switch_core_session_locate(m_sessionId.c_str());
@@ -131,7 +132,7 @@ public:
 
 				// send any buffered audio
 				int nFrames = m_audioBuffer.getNumItems();
-				switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p got stream ready, %d buffered frames\n", this, nFrames);	
+				switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p got stream ready, %d buffered frames\n", this, nFrames);
 				if (nFrames) {
 					char *p;
 					do {
@@ -141,19 +142,19 @@ public:
 						}
 					} while (p);
 				}
-	
+
 				switch_core_session_rwunlock(psession);
 			}
     };
-    auto OnResponseCallback = [this](const TranscribeStreamingServiceClient* pClient, 
-			const Model::StartStreamTranscriptionRequest& request, 
-			const Model::StartStreamTranscriptionOutcome& outcome, 
+    auto OnResponseCallback = [this](const TranscribeStreamingServiceClient* pClient,
+			const Model::StartStreamTranscriptionRequest& request,
+			const Model::StartStreamTranscriptionOutcome& outcome,
 			const std::shared_ptr<const Aws::Client::AsyncCallerContext>& context)
     {
  			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p stream got final response\n", this);
 			switch_core_session_t* psession = switch_core_session_locate(m_sessionId.c_str());
 			if (psession) {
-				if (!outcome.IsSuccess()) {				
+				if (!outcome.IsSuccess()) {
 					const TranscribeStreamingServiceError& err = outcome.GetError();
 					auto message = err.GetMessage();
 					auto exception = err.GetExceptionName();
@@ -185,7 +186,7 @@ public:
 
 
 	~GStreamer() {
-		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer::~GStreamer wrote %u packets %p\n", m_packets, this);		
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer::~GStreamer wrote %u packets %p\n", m_packets, this);
 	}
 
 	bool write(void* data, uint32_t datalen) {
@@ -227,7 +228,7 @@ public:
 		bool shutdownInitiated = false;
 		while (true) {
 			std::unique_lock<std::mutex> lk(m_mutex);
-			m_cond.wait(lk, [&, this] { 
+			m_cond.wait(lk, [&, this] {
 				return (!m_deqAudio.empty() && !m_finishing)  || m_transcript.TranscriptHasBeenSet() || m_finished  || (m_finishing && !shutdownInitiated);
 			});
 
@@ -263,7 +264,7 @@ public:
 						m_responseHandler(psession, s.str().c_str(), m_bugname.c_str());
 					}
 					TranscriptEvent empty;
-					m_transcript = empty; 
+					m_transcript = empty;
 
 					switch_core_session_rwunlock(psession);
 				}
@@ -320,8 +321,8 @@ static void *SWITCH_THREAD_FUNC aws_transcribe_thread(switch_thread_t *thread, v
 	struct cap_cb *cb = (struct cap_cb *) obj;
 	bool ok = true;
 	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "transcribe_thread: starting cb %p\n", (void *) cb);
-	GStreamer* pStreamer = new GStreamer(cb->sessionId, cb->bugname, cb->channels, cb->lang, cb->interim, cb->samples_per_second, cb->region, cb->awsAccessKeyId, cb->awsSecretAccessKey, 
-		cb->responseHandler);
+	GStreamer* pStreamer = new GStreamer(cb->sessionId, cb->bugname, cb->channels, cb->lang, cb->interim, cb->samples_per_second,
+		cb->region, cb->awsAccessKeyId, cb->awsSecretAccessKey, cb->awsSessionToken, cb->responseHandler);
 	if (!pStreamer) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "transcribe_thread: Error allocating streamer\n");
 		return nullptr;
@@ -361,7 +362,7 @@ extern "C" {
 		const char* secretAccessKey = std::getenv("AWS_SECRET_ACCESS_KEY");
 		const char* region = std::getenv("AWS_REGION");
 		if (NULL == accessKeyId && NULL == secretAccessKey) {
-			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, 
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE,
 				"\"AWS_ACCESS_KEY_ID\"  and/or \"AWS_SECRET_ACCESS_KEY\" env var not set; authentication will expect channel variables of same names to be set\n");
 		}
 		else {
@@ -369,7 +370,7 @@ extern "C" {
 
 		}
     Aws::SDKOptions options;
-/*		
+/*
     options.loggingOptions.logLevel = Aws::Utils::Logging::LogLevel::Trace;
 
 		Aws::Utils::Logging::InitializeAWSLogging(
@@ -380,7 +381,7 @@ extern "C" {
 
 		return SWITCH_STATUS_SUCCESS;
 	}
-	
+
 	switch_status_t aws_transcribe_cleanup() {
 		Aws::SDKOptions options;
 		/*
@@ -393,7 +394,7 @@ extern "C" {
 	}
 
 	// start transcribe on a channel
-	switch_status_t aws_transcribe_session_init(switch_core_session_t *session, responseHandler_t responseHandler, 
+	switch_status_t aws_transcribe_session_init(switch_core_session_t *session, responseHandler_t responseHandler,
           uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char* bugname, void **ppUserData
 	) {
 		switch_status_t status = SWITCH_STATUS_SUCCESS;
@@ -408,6 +409,7 @@ extern "C" {
 		memset(cb, sizeof(cb), 0);
 		const char* awsAccessKeyId = switch_channel_get_variable(channel, "AWS_ACCESS_KEY_ID");
 		const char* awsSecretAccessKey = switch_channel_get_variable(channel, "AWS_SECRET_ACCESS_KEY");
+		const char* awsSessionToken = switch_channel_get_variable(channel, "AWS_SESSION_TOKEN");
 		const char* awsRegion = switch_channel_get_variable(channel, "AWS_REGION");
 		cb->channels = channels;
 		LanguageCode code = LanguageCodeMapper::GetLanguageCodeForName(lang);
@@ -419,19 +421,19 @@ extern "C" {
 		strncpy(cb->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
 		strncpy(cb->bugname, bugname, MAX_BUG_LEN);
 
-		if (awsAccessKeyId && awsSecretAccessKey && awsRegion) {
+    if (awsRegion) strncpy(cb->region, awsRegion, MAX_REGION);
+		if (awsAccessKeyId && awsSecretAccessKey) {
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "Using channel vars for aws authentication\n");
 			strncpy(cb->awsAccessKeyId, awsAccessKeyId, 128);
 			strncpy(cb->awsSecretAccessKey, awsSecretAccessKey, 128);
-			strncpy(cb->region, awsRegion, MAX_REGION);
-
+			if (awsSessionToken) strncpy(cb->awsSessionToken, awsSessionToken, 1024);
 		}
 		else if (std::getenv("AWS_ACCESS_KEY_ID") &&
 			std::getenv("AWS_SECRET_ACCESS_KEY") &&
 			std::getenv("AWS_REGION")) {
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "Using env vars for aws authentication\n");
 			strncpy(cb->awsAccessKeyId, std::getenv("AWS_ACCESS_KEY_ID"), 128);
-			strncpy(cb->awsSecretAccessKey, std::getenv("AWS_SECRET_ACCESS_KEY"), 128);		
+			strncpy(cb->awsSecretAccessKey, std::getenv("AWS_SECRET_ACCESS_KEY"), 128);
 			strncpy(cb->region, std::getenv("AWS_REGION"), MAX_REGION);
 		}
 		else {
@@ -443,7 +445,7 @@ extern "C" {
 		if (switch_mutex_init(&cb->mutex, SWITCH_MUTEX_NESTED, pool) != SWITCH_STATUS_SUCCESS) {
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing mutex\n");
 			status = SWITCH_STATUS_FALSE;
-			goto done; 
+			goto done;
 		}
 
 		cb->interim = interim;
@@ -453,7 +455,7 @@ extern "C" {
 		if (sampleRate != 8000) {
 			cb->resampler = speex_resampler_init(1, sampleRate, 16000, SWITCH_RESAMPLE_QUALITY, &err);
 			if (0 != err) {
-				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing resampler: %s.\n", 
+				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing resampler: %s.\n",
 							switch_channel_get_name(channel), speex_resampler_strerror(err));
 				status = SWITCH_STATUS_FALSE;
 				goto done;
@@ -486,7 +488,7 @@ extern "C" {
 				switch_vad_set_param(cb->vad, "silence_ms", silence_ms);
 				switch_vad_set_param(cb->vad, "voice_ms", voice_ms);
 				switch_vad_set_param(cb->vad, "debug", debug);
-				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "%s: delaying connection until vad, voice_ms %d, mode %d\n", 
+				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "%s: delaying connection until vad, voice_ms %d, mode %d\n",
 					switch_channel_get_name(channel), voice_ms, mode);
 			}
 		}
@@ -497,7 +499,7 @@ extern "C" {
 		switch_thread_create(&cb->thread, thd_attr, aws_transcribe_thread, cb, pool);
 
 		*ppUserData = cb;
-	
+
 	done:
 		return status;
 	}
@@ -512,7 +514,18 @@ extern "C" {
 
 			// close connection and get final responses
 			switch_mutex_lock(cb->mutex);
-			GStreamer* streamer = (GStreamer *) cb->streamer;
+			GStreamer* streamer;
+
+      // race condition: if we just started a transcribe at same instant, the spawned thread may not have had a chance to set the streamer
+      int i = 0;
+      do {
+        streamer = (GStreamer *) cb->streamer;
+        if (streamer) break;
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG,
+          "aws_transcribe_session_stop: waiting for streamer to come online..%s\n", bugname);
+        switch_yield(10000); // wait 10ms
+      } while (i++ < 3);
+
 			if (streamer) {
 				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "aws_transcribe_session_stop: finish..%s\n", bugname);
 				streamer->finish();
@@ -544,7 +557,7 @@ extern "C" {
 		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s Bug is not attached.\n", switch_channel_get_name(channel));
 		return SWITCH_STATUS_FALSE;
 	}
-	
+
 	switch_bool_t aws_transcribe_frame(switch_media_bug_t *bug, void* user_data) {
 		switch_core_session_t *session = switch_core_media_bug_get_session(bug);
 		uint8_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
@@ -574,7 +587,7 @@ extern "C" {
 						}
 
 						if (cb->resampler) {
-							speex_resampler_process_interleaved_int(cb->resampler, (const spx_int16_t *) frame.data, (spx_uint32_t *) &in_len, &out[0], &out_len);						
+							speex_resampler_process_interleaved_int(cb->resampler, (const spx_int16_t *) frame.data, (spx_uint32_t *) &in_len, &out[0], &out_len);
 							streamer->write( &out[0], sizeof(spx_int16_t) * out_len);
 						}
 						else {
@@ -584,11 +597,11 @@ extern "C" {
 				}
 			}
 			else {
-				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, 
+				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG,
 					"aws_transcribe_frame: not sending audio because aws channel has been closed\n");
 			}
 			switch_mutex_unlock(cb->mutex);
 		}
 		return SWITCH_TRUE;
 	}
-}
+}

mod_aws_transcribe/mod_aws_transcribe.h

@@ -28,6 +28,7 @@ struct cap_cb {
 	char sessionId[MAX_SESSION_ID+1];
   char awsAccessKeyId[128];
   char awsSecretAccessKey[128];
+	char awsSessionToken[1024];
 	uint32_t channels;
   SpeexResamplerState *resampler;
 	void* streamer;

mod_aws_transcribe_ws/Makefile.am

@@ -0,0 +1,10 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_aws_transcribe_ws
+
+mod_LTLIBRARIES = mod_aws_transcribe_ws.la
+mod_aws_transcribe_ws_la_SOURCES  = mod_aws_transcribe_ws.c aws_transcribe_glue.cpp transcribe_manager.cpp audio_pipe.cpp
+mod_aws_transcribe_ws_la_CFLAGS   = $(AM_CFLAGS)
+mod_aws_transcribe_ws_la_CXXFLAGS = $(AM_CXXFLAGS) -std=c++11 -I${switch_srcdir}/libs/aws-sdk-cpp/aws-cpp-sdk-core/include -I${switch_srcdir}/libs/aws-sdk-cpp/aws-cpp-sdk-transcribestreaming/include -I${switch_srcdir}/libs/aws-sdk-cpp/build/.deps/install/include
+
+mod_aws_transcribe_ws_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_aws_transcribe_ws_la_LDFLAGS  = -avoid-version -module -no-undefined -shared `pkg-config --libs libwebsockets` 

mod_aws_transcribe_ws/README.md

@@ -0,0 +1,58 @@
+# mod_aws_transcribe
+
+A Freeswitch module that generates real-time transcriptions on a Freeswitch channel by using AWS streaming transcription API
+
+## API
+
+### Commands
+The freeswitch module exposes the following API commands:
+
+```
+aws_transcribe <uuid> start <lang-code> [interim]
+```
+Attaches media bug to channel and performs streaming recognize request.
+- `uuid` - unique identifier of Freeswitch channel
+- `lang-code` - a valid AWS [language code](https://docs.aws.amazon.com/transcribe/latest/dg/what-is-transcribe.html) that is supported for streaming transcription
+- `interim` - If the 'interim' keyword is present then both interim and final transcription results will be returned; otherwise only final transcriptions will be returned
+
+```
+aws_transcribe <uuid> stop
+```
+Stop transcription on the channel.
+
+### Authentication
+The plugin will first look for channel variables, then environment variables.  If neither are found, then the default AWS profile on the server will be used.
+
+The names of the channel variables and environment variables are:
+
+| variable | Description |
+| --- | ----------- |
+| AWS_ACCESS_KEY_ID | The Aws access key ID |
+| AWS_SECRET_ACCESS_KEY | The Aws secret access key |
+| AWS_REGION | The Aws region |
+
+
+### Events
+`aws_transcribe::transcription` - returns an interim or final transcription.  The event contains a JSON body describing the transcription result:
+```js
+[
+  {
+    "is_final": true,
+    "alternatives": [{
+      "transcript": "Hello. Can you hear me?"
+    }]
+  }
+]
+```
+
+## Usage
+When using [drachtio-fsrmf](https://www.npmjs.com/package/drachtio-fsmrf), you can access this API command via the api method on the 'endpoint' object.
+```js
+ep.api('aws_transcribe', `${ep.uuid} start en-US interim`);  
+```
+
+## Building
+This uses the AWS websocket api.
+
+## Examples
+[aws_transcribe.js](../../examples/aws_transcribe.js)

mod_aws_transcribe_ws/audio_pipe.cpp

@@ -0,0 +1,612 @@
+#include "audio_pipe.hpp"
+#include "transcribe_manager.hpp"
+#include "crc.h"
+
+#include <switch.h>
+
+#include <cassert>
+#include <iostream>
+#include <netinet/in.h>
+
+#include <fstream>
+#include <string>
+
+/* discard incoming text messages over the socket that are longer than this */
+#define MAX_RECV_BUF_SIZE (65 * 1024 * 10)
+#define RECV_BUF_REALLOC_SIZE (8 * 1024)
+#define AWS_PRELUDE_PLUS_HDRS_LEN (100)
+
+using namespace aws;
+
+namespace {
+  static const char *requestedTcpKeepaliveSecs = std::getenv("MOD_AUDIO_FORK_TCP_KEEPALIVE_SECS");
+  static int nTcpKeepaliveSecs = requestedTcpKeepaliveSecs ? ::atoi(requestedTcpKeepaliveSecs) : 55;
+  static uint8_t aws_prelude_and_headers[AWS_PRELUDE_PLUS_HDRS_LEN];
+
+  void writeToFile(const char* buffer, size_t bufferSize) {
+    static int writeCounter = 0; // Static variable to keep track of write count
+
+    // Write only the first three times
+    if (writeCounter >= 4) {
+        return;
+    }
+
+    // Generate a unique file name using the writeCounter
+    std::string filename = "/tmp/audio_data_" + std::to_string(writeCounter) + ".bin";
+
+    // Open a file in binary mode
+    std::ofstream outFile(filename, std::ios::binary);
+
+    // Check if the file is open
+    if (outFile.is_open()) {
+        // Write the buffer to the file
+        outFile.write(buffer, bufferSize);
+        outFile.close();
+
+        // Increment the write counter
+        writeCounter++;
+    } else {
+        // Handle error in file opening
+        std::cerr << "Unable to open file: " << filename << std::endl;
+    }
+  }
+}
+
+
+int AudioPipe::aws_lws_callback(struct lws *wsi, 
+  enum lws_callback_reasons reason,
+  void *user, void *in, size_t len) {
+
+  struct AudioPipe::lws_per_vhost_data *vhd = 
+    (struct AudioPipe::lws_per_vhost_data *) lws_protocol_vh_priv_get(lws_get_vhost(wsi), lws_get_protocol(wsi));
+
+  struct lws_vhost* vhost = lws_get_vhost(wsi);
+  AudioPipe ** ppAp = (AudioPipe **) user;
+
+  switch (reason) {
+    case LWS_CALLBACK_PROTOCOL_INIT:
+      vhd = (struct AudioPipe::lws_per_vhost_data *) lws_protocol_vh_priv_zalloc(lws_get_vhost(wsi), lws_get_protocol(wsi), sizeof(struct AudioPipe::lws_per_vhost_data));
+      vhd->context = lws_get_context(wsi);
+      vhd->protocol = lws_get_protocol(wsi);
+      vhd->vhost = lws_get_vhost(wsi);
+      break;
+
+    case LWS_CALLBACK_EVENT_WAIT_CANCELLED:
+      processPendingConnects(vhd);
+      processPendingDisconnects(vhd);
+      processPendingWrites();
+      break;
+    case LWS_CALLBACK_CLIENT_CONNECTION_ERROR:
+      {
+        AudioPipe* ap = findAndRemovePendingConnect(wsi);
+        int rc = lws_http_client_http_response(wsi);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR: %s, response status %d\n", in ? (char *)in : "(null)", rc); 
+        if (ap) {
+          ap->m_state = LWS_CLIENT_FAILED;
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_FAIL, (char *) in, ap->isFinished());
+        }
+        else {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_ESTABLISHED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+        }
+      }      
+      break;
+
+    case LWS_CALLBACK_CLIENT_ESTABLISHED:
+      {
+        AudioPipe* ap = findAndRemovePendingConnect(wsi);
+        if (ap) {
+          *ppAp = ap;
+          ap->m_vhd = vhd;
+          ap->m_state = LWS_CLIENT_CONNECTED;
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_SUCCESS, NULL,  ap->isFinished());
+          //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"%s connected\n", ap->m_uuid.c_str());
+        }
+        else {
+          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_ESTABLISHED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+        }
+      }      
+      break;
+    case LWS_CALLBACK_CLIENT_CLOSED:
+      {
+        AudioPipe* ap = *ppAp;
+        if (!ap) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CLOSED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          return 0;
+        }
+        if (ap->m_state == LWS_CLIENT_DISCONNECTING) {
+          // closed by us
+
+          lwsl_debug("%s socket closed by us\n", ap->m_uuid.c_str());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_CLOSED_GRACEFULLY, NULL,  ap->isFinished());
+        }
+        else if (ap->m_state == LWS_CLIENT_CONNECTED) {
+          // closed by far end
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"%s socket closed by far end\n", ap->m_uuid.c_str());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_DROPPED, NULL,  ap->isFinished());
+        }
+        ap->m_state = LWS_CLIENT_DISCONNECTED;
+        ap->setClosed();
+    
+        //NB: after receiving any of the events above, any holder of a 
+        //pointer or reference to this object must treat is as no longer valid
+
+        //*ppAp = NULL;
+        //delete ap;
+      }
+      break;
+
+    case LWS_CALLBACK_CLIENT_RECEIVE:
+      {
+        AudioPipe* ap = *ppAp;
+        if (!ap) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          return 0;
+        }
+
+        if (lws_is_first_fragment(wsi)) {
+          // allocate a buffer for the entire chunk of memory needed
+          assert(nullptr == ap->m_recv_buf);
+          ap->m_recv_buf_len = len + lws_remaining_packet_payload(wsi);
+          ap->m_recv_buf = (uint8_t*) malloc(ap->m_recv_buf_len);
+          ap->m_recv_buf_ptr = ap->m_recv_buf;
+        }
+
+        size_t write_offset = ap->m_recv_buf_ptr - ap->m_recv_buf;
+        size_t remaining_space = ap->m_recv_buf_len - write_offset;
+        if (remaining_space < len) {
+          lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE buffer realloc needed.\n");
+          size_t newlen = ap->m_recv_buf_len + RECV_BUF_REALLOC_SIZE;
+          if (newlen > MAX_RECV_BUF_SIZE) {
+            free(ap->m_recv_buf);
+            ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
+            ap->m_recv_buf_len = 0;
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE max buffer exceeded, truncating message.\n");
+          }
+          else {
+            ap->m_recv_buf = (uint8_t*) realloc(ap->m_recv_buf, newlen);
+            if (nullptr != ap->m_recv_buf) {
+              ap->m_recv_buf_len = newlen;
+              ap->m_recv_buf_ptr = ap->m_recv_buf + write_offset;
+            }
+          }
+        }
+
+        if (nullptr != ap->m_recv_buf) {
+          if (len > 0) {
+            memcpy(ap->m_recv_buf_ptr, in, len);
+            ap->m_recv_buf_ptr += len;
+          }
+          if (lws_is_final_fragment(wsi)) {
+            //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "AudioPipe::lws_service_thread - LWS_CALLBACK_CLIENT_RECEIVE received %d bytes.\n", len);
+            if (nullptr != ap->m_recv_buf) {
+              bool isError = false;
+              std::string payload;
+              std::string msg((char *)ap->m_recv_buf, ap->m_recv_buf_ptr - ap->m_recv_buf);
+
+              TranscribeManager::parseResponse(msg, payload, isError, true);
+              //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE payload: %s.\n", payload.c_str());
+              //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE response %s.\n", msg.c_str());
+
+
+              if (0 != payload.compare("{\"Transcript\":{\"Results\":[]}}")) {
+                ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::MESSAGE, payload.c_str(),  ap->isFinished());
+              }
+              if (nullptr != ap->m_recv_buf) free(ap->m_recv_buf);
+            }
+            ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
+            ap->m_recv_buf_len = 0;
+          }
+        }
+      }
+      break;
+
+    case LWS_CALLBACK_CLIENT_WRITEABLE:
+      {
+        AudioPipe* ap = *ppAp;
+        if (!ap) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_WRITEABLE %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
+          return 0;
+        }
+
+        if (ap->m_state == LWS_CLIENT_DISCONNECTING) {
+          lws_close_reason(wsi, LWS_CLOSE_STATUS_NORMAL, NULL, 0);
+          return -1;
+        }
+
+        // check for audio packets
+        {
+          std::lock_guard<std::mutex> lk(ap->m_audio_mutex);
+          if (ap->m_audio_buffer_write_offset > LWS_PRE + AWS_PRELUDE_PLUS_HDRS_LEN || ap->isFinished()) {
+
+            // send a zero length audio packet to indicate end of stream
+            if (ap->isFinished()) {
+              ap->m_audio_buffer_write_offset = LWS_PRE + AWS_PRELUDE_PLUS_HDRS_LEN;
+            }
+            /**
+             * fill in
+             * [0..3] = total byte length
+             * [8..11] = prelude crc
+             * following the audio data: 4 bytes of Message CRC
+             * 
+             */
+
+            // copy in the prelude and headers
+            memcpy(ap->m_audio_buffer + LWS_PRE, aws_prelude_and_headers, AWS_PRELUDE_PLUS_HDRS_LEN);
+
+            // fill in the total byte length
+            uint32_t totalLen = ap->m_audio_buffer_write_offset - LWS_PRE + 4; // for the trailing Message CRC which is 4 bytes
+            //lwsl_err("AudioPipe - total length %u (decimal), 0x%X (hex)\n", totalLen, totalLen);
+            totalLen = htonl(totalLen);
+            //lwsl_err("AudioPipe - total length in network byte order %u (decimal), 0x%X (hex)\n", totalLen, totalLen);
+            memcpy(ap->m_audio_buffer + LWS_PRE, &totalLen, sizeof(uint32_t));
+
+            // fill in the prelude crc
+            uint32_t preludeCRC = CRC::Calculate(ap->m_audio_buffer + LWS_PRE, 8, CRC::CRC_32());
+            //lwsl_err("AudioPipe - prelude CRC %u (decimal), 0x%X (hex)\n", preludeCRC, preludeCRC);
+            preludeCRC = htonl(preludeCRC);
+            //lwsl_err("AudioPipe - prelude CRC in network order %u (decimal), 0x%X (hex)\n", preludeCRC, preludeCRC);
+            memcpy(ap->m_audio_buffer + LWS_PRE + 8, &preludeCRC, sizeof(uint32_t));
+
+            // fill in the message crc
+            uint32_t messageCRC = CRC::Calculate(ap->m_audio_buffer + LWS_PRE, ap->m_audio_buffer_write_offset - LWS_PRE, CRC::CRC_32());
+            messageCRC = htonl(messageCRC);
+            memcpy(ap->m_audio_buffer + ap->m_audio_buffer_write_offset, &messageCRC, sizeof(uint32_t));
+            ap->m_audio_buffer_write_offset + sizeof(uint32_t);
+
+            size_t datalen = ap->m_audio_buffer_write_offset - LWS_PRE + 4;
+            
+            //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"%s writing data length %lu\n", ap->m_uuid.c_str(), datalen);
+
+            // TMP: write data to a file
+            //writeToFile((const char *) ap->m_audio_buffer + LWS_PRE, datalen);
+
+            int sent = lws_write(wsi, (unsigned char *) ap->m_audio_buffer + LWS_PRE, datalen, LWS_WRITE_BINARY);
+            if (sent < datalen) {
+              switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_WRITEABLE %s attemped to send %lu only sent %d wsi %p..\n", 
+                ap->m_uuid.c_str(), datalen, sent, wsi); 
+            }
+            ap->binaryWritePtrResetToZero();
+          }
+        }
+
+        return 0;
+      }
+      break;
+
+    default:
+      break;
+  }
+  return lws_callback_http_dummy(wsi, reason, user, in, len);
+}
+
+
+// static members
+static const lws_retry_bo_t retry = {
+    nullptr,   // retry_ms_table
+    0,         // retry_ms_table_count
+    0,         // conceal_count
+    UINT16_MAX,         // secs_since_valid_ping
+    UINT16_MAX,        // secs_since_valid_hangup
+    0          // jitter_percent
+};
+
+struct lws_context *AudioPipe::contexts[] = {
+  nullptr, nullptr, nullptr, nullptr, nullptr,
+  nullptr, nullptr, nullptr, nullptr, nullptr
+};
+unsigned int AudioPipe::numContexts = 0;
+unsigned int AudioPipe::nchild = 0;
+std::string AudioPipe::protocolName;
+std::mutex AudioPipe::mutex_connects;
+std::mutex AudioPipe::mutex_disconnects;
+std::mutex AudioPipe::mutex_writes;
+std::list<AudioPipe*> AudioPipe::pendingConnects;
+std::list<AudioPipe*> AudioPipe::pendingDisconnects;
+std::list<AudioPipe*> AudioPipe::pendingWrites;
+AudioPipe::log_emit_function AudioPipe::logger;
+std::mutex AudioPipe::mapMutex;
+std::unordered_map<std::thread::id, bool> AudioPipe::stopFlags;
+std::queue<std::thread::id> AudioPipe::threadIds;
+
+
+void AudioPipe::processPendingConnects(lws_per_vhost_data *vhd) {
+  std::list<AudioPipe*> connects;
+  {
+    std::lock_guard<std::mutex> guard(mutex_connects);
+    for (auto it = pendingConnects.begin(); it != pendingConnects.end(); ++it) {
+      if ((*it)->m_state == LWS_CLIENT_IDLE) {
+        connects.push_back(*it);
+        (*it)->m_state = LWS_CLIENT_CONNECTING;
+      }
+    }
+  }
+  for (auto it = connects.begin(); it != connects.end(); ++it) {
+    AudioPipe* ap = *it;
+    ap->connect_client(vhd);   
+  }
+}
+
+void AudioPipe::processPendingDisconnects(lws_per_vhost_data *vhd) {
+  std::list<AudioPipe*> disconnects;
+  {
+    std::lock_guard<std::mutex> guard(mutex_disconnects);
+    for (auto it = pendingDisconnects.begin(); it != pendingDisconnects.end(); ++it) {
+      if ((*it)->m_state == LWS_CLIENT_DISCONNECTING) disconnects.push_back(*it);
+    }
+    pendingDisconnects.clear();
+  }
+  for (auto it = disconnects.begin(); it != disconnects.end(); ++it) {
+    AudioPipe* ap = *it;
+    lws_callback_on_writable(ap->m_wsi); 
+  }
+}
+
+void AudioPipe::processPendingWrites() {
+  std::list<AudioPipe*> writes;
+  {
+    std::lock_guard<std::mutex> guard(mutex_writes);
+    for (auto it = pendingWrites.begin(); it != pendingWrites.end(); ++it) {
+       if ((*it)->m_state == LWS_CLIENT_CONNECTED) writes.push_back(*it);
+    }  
+    pendingWrites.clear();
+  }
+  for (auto it = writes.begin(); it != writes.end(); ++it) {
+    AudioPipe* ap = *it;
+    lws_callback_on_writable(ap->m_wsi);
+  }
+}
+
+AudioPipe* AudioPipe::findAndRemovePendingConnect(struct lws *wsi) {
+  AudioPipe* ap = NULL;
+  std::lock_guard<std::mutex> guard(mutex_connects);
+  std::list<AudioPipe* > toRemove;
+
+  for (auto it = pendingConnects.begin(); it != pendingConnects.end() && !ap; ++it) {
+    int state = (*it)->m_state;
+
+    if ((*it)->m_wsi == nullptr)
+      toRemove.push_back(*it);
+
+    if ((state == LWS_CLIENT_CONNECTING) &&
+      (*it)->m_wsi == wsi) ap = *it;
+  }
+
+  for (auto it = toRemove.begin(); it != toRemove.end(); ++it)
+    pendingConnects.remove(*it);
+
+  if (ap) {
+    pendingConnects.remove(ap);
+  }
+
+  return ap;
+}
+
+AudioPipe* AudioPipe::findPendingConnect(struct lws *wsi) {
+  AudioPipe* ap = NULL;
+  std::lock_guard<std::mutex> guard(mutex_connects);
+
+  for (auto it = pendingConnects.begin(); it != pendingConnects.end() && !ap; ++it) {
+    int state = (*it)->m_state;
+    if ((state == LWS_CLIENT_CONNECTING) &&
+      (*it)->m_wsi == wsi) ap = *it;
+  }
+  return ap;
+}
+
+void AudioPipe::addPendingConnect(AudioPipe* ap) {
+  {
+    std::lock_guard<std::mutex> guard(mutex_connects);
+    pendingConnects.push_back(ap);
+    lwsl_debug("%s after adding connect there are %lu pending connects\n", 
+      ap->m_uuid.c_str(), pendingConnects.size());
+  }
+  lws_cancel_service(contexts[nchild++ % numContexts]);
+}
+void AudioPipe::addPendingDisconnect(AudioPipe* ap) {
+  ap->m_state = LWS_CLIENT_DISCONNECTING;
+  {
+    std::lock_guard<std::mutex> guard(mutex_disconnects);
+    pendingDisconnects.push_back(ap);
+    lwsl_debug("%s after adding disconnect there are %lu pending disconnects\n", 
+      ap->m_uuid.c_str(), pendingDisconnects.size());
+  }
+  lws_cancel_service(ap->m_vhd->context);
+}
+void AudioPipe::addPendingWrite(AudioPipe* ap) {
+  {
+    std::lock_guard<std::mutex> guard(mutex_writes);
+    pendingWrites.push_back(ap);
+  }
+  lws_cancel_service(ap->m_vhd->context);
+}
+
+void AudioPipe::binaryWritePtrResetToZero(void) {
+    m_audio_buffer_write_offset = LWS_PRE + AWS_PRELUDE_PLUS_HDRS_LEN;
+}
+
+bool AudioPipe::lws_service_thread(unsigned int nServiceThread) {
+  struct lws_context_creation_info info;
+  std::thread::id this_id = std::this_thread::get_id();
+
+  const struct lws_protocols protocols[] = {
+    {
+      "",
+      AudioPipe::aws_lws_callback,
+      sizeof(void *),
+      1024,
+    },
+    { NULL, NULL, 0, 0 }
+  };
+
+  memset(&info, 0, sizeof info); 
+  info.port = CONTEXT_PORT_NO_LISTEN; 
+  info.options = LWS_SERVER_OPTION_DO_SSL_GLOBAL_INIT;
+  info.protocols = protocols;
+  info.ka_time = nTcpKeepaliveSecs;                    // tcp keep-alive timer
+  info.ka_probes = 4;                   // number of times to try ka before closing connection
+  info.ka_interval = 5;                 // time between ka's
+  info.timeout_secs = 10;                // doc says timeout for "various processes involving network roundtrips"
+  info.keepalive_timeout = 5;           // seconds to allow remote client to hold on to an idle HTTP/1.1 connection 
+  info.timeout_secs_ah_idle = 10;       // secs to allow a client to hold an ah without using it
+  info.retry_and_idle_policy = &retry;
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"AudioPipe::lws_service_thread creating context\n");
+
+  contexts[nServiceThread] = lws_create_context(&info);
+  if (!contexts[nServiceThread]) {
+    lwsl_err("AudioPipe::lws_service_thread failed creating context in service thread %d..\n", nServiceThread); 
+    return false;
+  }
+
+  int n;
+  do {
+    n = lws_service(contexts[nServiceThread], 0);
+  } while (n >= 0 && !stopFlags[this_id]);
+
+  // Cleanup once work is done or stopped
+  {
+      std::lock_guard<std::mutex> lock(mapMutex);
+      stopFlags.erase(this_id);
+  }
+
+  lwsl_notice("AudioPipe::lws_service_thread ending in service thread %d\n", nServiceThread); 
+  return true;
+}
+
+void AudioPipe::initialize(unsigned int nThreads, int loglevel, log_emit_function logger) {
+  assert(nThreads > 0 && nThreads <= 10);
+
+  numContexts = nThreads;
+  lws_set_log_level(loglevel, logger);
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE,"AudioPipe::initialize starting\n"); 
+  for (unsigned int i = 0; i < numContexts; i++) {
+    std::lock_guard<std::mutex> lock(mapMutex);
+    std::thread t(&AudioPipe::lws_service_thread, i);
+    stopFlags[t.get_id()] = false;
+    threadIds.push(t.get_id());
+    t.detach();
+  }
+}
+
+bool AudioPipe::deinitialize() {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE,"AudioPipe::deinitialize\n"); 
+  std::lock_guard<std::mutex> lock(mapMutex);
+  if (!threadIds.empty()) {
+      std::thread::id id = threadIds.front();
+      threadIds.pop();
+      stopFlags[id] = true;
+  }
+  for (unsigned int i = 0; i < numContexts; i++)
+  {
+    lwsl_notice("AudioPipe::deinitialize destroying context %d of %d\n", i + 1, numContexts);
+    lws_context_destroy(contexts[i]);
+  }
+  std::this_thread::sleep_for(std::chrono::seconds(2));
+  return true;
+}
+
+// instance members
+AudioPipe::AudioPipe(const char* uuid, const char* bugname, const char* host, unsigned int port, const char* path,
+  size_t bufLen, size_t minFreespace, notifyHandler_t callback) :
+  m_uuid(uuid), m_host(host), m_port(port), m_path(path), m_finished(false), m_bugname(bugname),
+  m_audio_buffer_min_freespace(minFreespace), m_audio_buffer_max_len(bufLen), m_gracefulShutdown(false),
+  m_recv_buf(nullptr), m_recv_buf_ptr(nullptr),
+  m_state(LWS_CLIENT_IDLE), m_wsi(nullptr), m_vhd(nullptr), m_callback(callback) {
+
+  char headerBuffer[88];
+  char* buffer = headerBuffer;
+  m_audio_buffer = new uint8_t[m_audio_buffer_max_len];
+
+  // stamp out the template for the prelude and headers 
+  memset(aws_prelude_and_headers, 0, AWS_PRELUDE_PLUS_HDRS_LEN);
+
+  // aws_prelude_and_headers[0..3] = total byte length (not known till message send time)
+
+  // aws_prelude_and_headers[4..7] = headers byte length
+  uint32_t headerLen = sizeof(headerBuffer);
+  headerLen = htonl(headerLen);
+  memcpy(&aws_prelude_and_headers[4], &headerLen, sizeof(uint32_t));
+
+  // aws_prelude_and_headers[8..11] = prelude crc (not known till message send time)
+
+  // aws_prelude_and_headers[12..99] = headers
+  TranscribeManager::writeHeader(&buffer, ":content-type", "application/octet-stream");
+  TranscribeManager::writeHeader(&buffer, ":event-type", "AudioEvent");
+  TranscribeManager::writeHeader(&buffer, ":message-type", "event");
+
+  memcpy(&aws_prelude_and_headers[12], headerBuffer, sizeof(headerBuffer));
+
+  // following this will be the audio data and a final message CRC (not known till message send time)
+
+  memcpy(m_audio_buffer + LWS_PRE, aws_prelude_and_headers, AWS_PRELUDE_PLUS_HDRS_LEN);
+  m_audio_buffer_write_offset = LWS_PRE + AWS_PRELUDE_PLUS_HDRS_LEN;
+
+  //writeToFile((const char *) m_audio_buffer + LWS_PRE, AWS_PRELUDE_PLUS_HDRS_LEN);
+
+}
+
+AudioPipe::~AudioPipe() {
+  if (m_audio_buffer) delete [] m_audio_buffer;
+  if (m_recv_buf) delete [] m_recv_buf;
+}
+
+void AudioPipe::connect(void) {
+  addPendingConnect(this);
+}
+
+bool AudioPipe::connect_client(struct lws_per_vhost_data *vhd) {
+  assert(m_audio_buffer != nullptr);
+  assert(m_vhd == nullptr);
+  struct lws_client_connect_info i;
+
+  memset(&i, 0, sizeof(i));
+  i.context = vhd->context;
+  i.port = m_port;
+  i.address = m_host.c_str();
+  i.path = m_path.c_str();
+  i.host = i.address;
+  i.origin = i.address;
+  i.ssl_connection = LCCSCF_USE_SSL;
+  //i.protocol = protocolName.c_str();
+  i.pwsi = &(m_wsi);
+
+  m_state = LWS_CLIENT_CONNECTING;
+  m_vhd = vhd;
+
+  m_wsi = lws_client_connect_via_info(&i);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG,"%s attempting connection, wsi is %p\n", m_uuid.c_str(), m_wsi);
+
+  return nullptr != m_wsi;
+}
+
+void AudioPipe::bufferForSending(const char* text) {
+  if (m_state != LWS_CLIENT_CONNECTED) return;
+  {
+    std::lock_guard<std::mutex> lk(m_text_mutex);
+    m_metadata.append(text);
+  }
+  addPendingWrite(this);
+}
+
+void AudioPipe::unlockAudioBuffer() {
+  if (m_audio_buffer_write_offset > LWS_PRE) addPendingWrite(this);
+  m_audio_mutex.unlock();
+}
+
+void AudioPipe::close() {
+  if (m_state != LWS_CLIENT_CONNECTED) return;
+  addPendingDisconnect(this);
+}
+
+void AudioPipe::finish() {
+  if (m_finished || m_state != LWS_CLIENT_CONNECTED) return;
+  m_finished = true;
+  addPendingWrite(this);
+}
+
+void AudioPipe::waitForClose() {
+  std::shared_future<void> sf(m_promise.get_future());
+  sf.wait();
+  return;
+}

mod_aws_transcribe_ws/audio_pipe.hpp

@@ -0,0 +1,144 @@
+#ifndef __AWS_AUDIO_PIPE_HPP__
+#define __AWS_AUDIO_PIPE_HPP__
+
+#include <string>
+#include <list>
+#include <mutex>
+#include <future>
+#include <queue>
+#include <unordered_map>
+#include <thread>
+
+#include <libwebsockets.h>
+
+namespace aws {
+
+class AudioPipe {
+public:
+  enum LwsState_t {
+    LWS_CLIENT_IDLE,
+    LWS_CLIENT_CONNECTING,
+    LWS_CLIENT_CONNECTED,
+    LWS_CLIENT_FAILED,
+    LWS_CLIENT_DISCONNECTING,
+    LWS_CLIENT_DISCONNECTED
+  };
+  enum NotifyEvent_t {
+    CONNECT_SUCCESS,
+    CONNECT_FAIL,
+    CONNECTION_DROPPED,
+    CONNECTION_CLOSED_GRACEFULLY,
+    MESSAGE
+  };
+  typedef void (*log_emit_function)(int level, const char *line);
+  typedef void (*notifyHandler_t)(const char *sessionId, const char* bugname, NotifyEvent_t event, const char* message, bool finished);
+
+  struct lws_per_vhost_data {
+    struct lws_context *context;
+    struct lws_vhost *vhost;
+    const struct lws_protocols *protocol;
+  };
+
+  static void initialize(unsigned int nThreads, int loglevel, log_emit_function logger);
+  static bool deinitialize();
+  static bool lws_service_thread(unsigned int nServiceThread);
+
+  // constructor
+  AudioPipe(const char* uuid, const char* bugname, const char* host, unsigned int port, const char* path, 
+    size_t bufLen, size_t minFreespace, notifyHandler_t callback);
+  ~AudioPipe();  
+
+  LwsState_t getLwsState(void) { return m_state; }
+  std::string& getApiKey(void) {
+    return m_apiKey;
+  }
+  void connect(void);
+  void bufferForSending(const char* text);
+  size_t binarySpaceAvailable(void) {
+    return m_audio_buffer_max_len - m_audio_buffer_write_offset;
+  }
+  size_t binaryMinSpace(void) {
+    return m_audio_buffer_min_freespace;
+  }
+  char * binaryWritePtr(void) { 
+    return (char *) m_audio_buffer + m_audio_buffer_write_offset;
+  }
+  void binaryWritePtrAdd(size_t len) {
+    m_audio_buffer_write_offset += len;
+  }
+  void binaryWritePtrResetToZero(void);
+  void lockAudioBuffer(void) {
+    m_audio_mutex.lock();
+  }
+  void unlockAudioBuffer(void) ;
+
+  void close() ;
+  void finish();
+  void waitForClose();
+  void setClosed() { m_promise.set_value(); }
+  bool isFinished() { return m_finished;}
+
+  // no default constructor or copying
+  AudioPipe() = delete;
+  AudioPipe(const AudioPipe&) = delete;
+  void operator=(const AudioPipe&) = delete;
+
+private:
+
+  static int aws_lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
+  static unsigned int nchild;
+  static struct lws_context *contexts[];
+  static unsigned int numContexts;
+  static std::string protocolName;
+  static std::mutex mutex_connects;
+  static std::mutex mutex_disconnects;
+  static std::mutex mutex_writes;
+  static std::list<AudioPipe*> pendingConnects;
+  static std::list<AudioPipe*> pendingDisconnects;
+  static std::list<AudioPipe*> pendingWrites;
+  static log_emit_function logger;
+
+  static std::mutex mapMutex;
+  static std::unordered_map<std::thread::id, bool> stopFlags;
+  static std::queue<std::thread::id> threadIds;
+
+  static AudioPipe* findAndRemovePendingConnect(struct lws *wsi);
+  static AudioPipe* findPendingConnect(struct lws *wsi);
+  static void addPendingConnect(AudioPipe* ap);
+  static void addPendingDisconnect(AudioPipe* ap);
+  static void addPendingWrite(AudioPipe* ap);
+  static void processPendingConnects(lws_per_vhost_data *vhd);
+  static void processPendingDisconnects(lws_per_vhost_data *vhd);
+  static void processPendingWrites(void);
+  
+  bool connect_client(struct lws_per_vhost_data *vhd);
+
+  LwsState_t m_state;
+  std::string m_uuid;
+  std::string m_host;
+  unsigned int m_port;
+  std::string m_path;
+  std::string m_metadata;
+  std::mutex m_text_mutex;
+  std::mutex m_audio_mutex;
+  int m_sslFlags;
+  struct lws *m_wsi;
+  uint8_t *m_audio_buffer;
+  size_t m_audio_buffer_max_len;
+  size_t m_audio_buffer_write_offset;
+  size_t m_audio_buffer_min_freespace;
+  uint8_t* m_recv_buf;
+  uint8_t* m_recv_buf_ptr;
+  size_t m_recv_buf_len;
+  struct lws_per_vhost_data* m_vhd;
+  notifyHandler_t m_callback;
+  log_emit_function m_logger;
+  std::string m_apiKey;
+  bool m_gracefulShutdown;
+  bool m_finished;
+  std::string m_bugname;
+  std::promise<void> m_promise;
+};
+
+} // namespace deepgram
+#endif

mod_aws_transcribe_ws/aws_transcribe_glue.cpp

@@ -0,0 +1,415 @@
+#include <switch.h>
+#include <switch_json.h>
+#include <string.h>
+#include <string>
+#include <mutex>
+#include <thread>
+#include <list>
+#include <algorithm>
+#include <functional>
+#include <cassert>
+#include <cstdlib>
+#include <fstream>
+#include <sstream>
+#include <regex>
+#include <iostream>
+#include <unordered_map>
+
+#include "mod_aws_transcribe_ws.h"
+#include "simple_buffer.h"
+//#include "parser.hpp"
+#include "audio_pipe.hpp"
+#include "transcribe_manager.hpp"
+
+#define RTP_PACKETIZATION_PERIOD 20
+#define FRAME_SIZE_8000  320 /*which means each 20ms frame as 320 bytes at 8 khz (1 channel only)*/
+
+namespace {
+  static bool hasDefaultCredentials = false;
+  static const char* defaultApiKey = nullptr;
+  static const char *requestedBufferSecs = std::getenv("MOD_AUDIO_FORK_BUFFER_SECS");
+  static int nAudioBufferSecs = std::max(1, std::min(requestedBufferSecs ? ::atoi(requestedBufferSecs) : 2, 5));
+  static const char *requestedNumServiceThreads = std::getenv("MOD_AUDIO_FORK_SERVICE_THREADS");
+  static unsigned int nServiceThreads = std::max(1, std::min(requestedNumServiceThreads ? ::atoi(requestedNumServiceThreads) : 1, 5));
+  static unsigned int idxCallCount = 0;
+  static uint32_t playCount = 0;
+
+  static const char* emptyTranscript = "{\"Transcript\":{\"Results\":[]}}";
+  static const char* messageStart = "{\"Message\":";
+
+  static void reaper(private_t *tech_pvt) {
+    std::shared_ptr<aws::AudioPipe> pAp;
+    pAp.reset((aws::AudioPipe *)tech_pvt->pAudioPipe);
+    tech_pvt->pAudioPipe = nullptr;
+
+    std::thread t([pAp, tech_pvt]{
+      pAp->finish();
+      pAp->waitForClose();
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "%s (%u) got remote close\n", tech_pvt->sessionId, tech_pvt->id);
+    });
+    t.detach();
+  }
+
+  static void destroy_tech_pvt(private_t *tech_pvt) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "%s (%u) destroy_tech_pvt\n", tech_pvt->sessionId, tech_pvt->id);
+    if (tech_pvt) {
+      if (tech_pvt->pAudioPipe) {
+        aws::AudioPipe* p = (aws::AudioPipe *) tech_pvt->pAudioPipe;
+        delete p;
+        tech_pvt->pAudioPipe = nullptr;
+      }
+      if (tech_pvt->resampler) {
+          speex_resampler_destroy(tech_pvt->resampler);
+          tech_pvt->resampler = NULL;
+      }
+
+      if (tech_pvt->vad) {
+        switch_vad_destroy(&tech_pvt->vad);
+        tech_pvt->vad = nullptr;
+      }
+    }
+  }
+
+  static void eventCallback(const char* sessionId, const char* bugname, 
+    aws::AudioPipe::NotifyEvent_t event, const char* message, bool finished) {
+    switch_core_session_t* session = switch_core_session_locate(sessionId);
+    if (session) {
+      switch_channel_t *channel = switch_core_session_get_channel(session);
+      switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, bugname);
+      if (bug) {
+        private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
+        if (tech_pvt) {
+          switch (event) {
+            case aws::AudioPipe::CONNECT_SUCCESS:
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "connection successful\n");
+              tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_CONNECT_SUCCESS, NULL, tech_pvt->bugname, finished);
+            break;
+            case aws::AudioPipe::CONNECT_FAIL:
+            {
+              // first thing: we can no longer access the AudioPipe
+              std::stringstream json;
+              json << "{\"reason\":\"" << message << "\"}";
+              tech_pvt->pAudioPipe = nullptr;
+              tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_CONNECT_FAIL, (char *) json.str().c_str(), tech_pvt->bugname, finished);
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_NOTICE, "connection failed: %s\n", message);
+            }
+            break;
+            case aws::AudioPipe::CONNECTION_DROPPED:
+              // first thing: we can no longer access the AudioPipe
+              tech_pvt->pAudioPipe = nullptr;
+              tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_DISCONNECT, NULL, tech_pvt->bugname, finished);
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection dropped from far end\n");
+            break;
+            case aws::AudioPipe::CONNECTION_CLOSED_GRACEFULLY:
+              // first thing: we can no longer access the AudioPipe
+              tech_pvt->pAudioPipe = nullptr;
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection closed gracefully\n");
+            break;
+            case aws::AudioPipe::MESSAGE:
+              if( strstr(message, emptyTranscript)) {
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "discarding empty aws transcript\n");
+              }
+              else if (0 == strncmp( message, messageStart, strlen(messageStart))) {
+                tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_ERROR, message, tech_pvt->bugname, finished);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "error message from aws: %s\n", message);
+              }
+              else {
+                tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_RESULTS, message, tech_pvt->bugname, finished);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "aws message: %s.\n", message);
+              }
+            break;
+
+            default:
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_NOTICE, "got unexpected msg from aws %d:%s\n", event, message);
+              break;
+          }
+        }
+      }
+      switch_core_session_rwunlock(session);
+    }
+  }
+
+  void lws_logger(int level, const char *line) {
+    switch_log_level_t llevel = SWITCH_LOG_DEBUG;
+
+    switch (level) {
+      case LLL_ERR: llevel = SWITCH_LOG_ERROR; break;
+      case LLL_WARN: llevel = SWITCH_LOG_WARNING; break;
+      case LLL_NOTICE: llevel = SWITCH_LOG_NOTICE; break;
+      case LLL_INFO: llevel = SWITCH_LOG_INFO; break;
+      break;
+    }
+	  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "%s\n", line);
+  }
+}
+
+
+extern "C" {
+  switch_status_t aws_transcribe_init() {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_aws_transcribe: audio buffer (in secs):    %d secs\n", nAudioBufferSecs);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_aws_transcribe: lws service threads:       %d\n", nServiceThreads);
+ 
+    int logs = LLL_ERR | LLL_WARN | LLL_NOTICE || LLL_INFO | LLL_PARSER | LLL_HEADER | LLL_EXT | LLL_CLIENT  | LLL_LATENCY | LLL_DEBUG ;
+    
+    aws::AudioPipe::initialize(nServiceThreads, logs, lws_logger);
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AudioPipe::initialize completed\n");
+
+		return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t aws_transcribe_cleanup() {
+    bool cleanup = false;
+    cleanup = aws::AudioPipe::deinitialize();
+    if (cleanup == true) {
+        return SWITCH_STATUS_SUCCESS;
+    }
+    return SWITCH_STATUS_FALSE;
+  }
+	
+  	// start transcribe on a channel
+	switch_status_t aws_transcribe_session_init(switch_core_session_t *session, responseHandler_t responseHandler, 
+          uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char* bugname, void **ppUserData
+	) {
+		switch_status_t status = SWITCH_STATUS_SUCCESS;
+		switch_channel_t *channel = switch_core_session_get_channel(session);
+		int err;
+    uint32_t desiredSampling = 8000;
+		switch_threadattr_t *thd_attr = NULL;
+		switch_memory_pool_t *pool = switch_core_session_get_pool(session);
+		auto read_codec = switch_core_session_get_read_codec(session);
+		uint32_t sampleRate = read_codec->implementation->actual_samples_per_second;
+    switch_codec_implementation_t read_impl;
+    switch_core_session_get_read_impl(session, &read_impl);
+
+		private_t* tech_pvt = (private_t *) switch_core_session_alloc(session, sizeof(private_t));
+		memset(tech_pvt, sizeof(tech_pvt), 0);
+		const char* awsAccessKeyId = switch_channel_get_variable(channel, "AWS_ACCESS_KEY_ID");
+		const char* awsSecretAccessKey = switch_channel_get_variable(channel, "AWS_SECRET_ACCESS_KEY");
+		const char* awsRegion = switch_channel_get_variable(channel, "AWS_REGION");
+		const char* awsSessionToken = switch_channel_get_variable(channel, "AWS_SECURITY_TOKEN");
+		tech_pvt->channels = channels;
+		strncpy(tech_pvt->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
+		strncpy(tech_pvt->bugname, bugname, MAX_BUG_LEN);
+
+		if (awsAccessKeyId && awsSecretAccessKey && awsRegion && awsSessionToken) {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "Using channel vars for aws authentication\n");
+			strncpy(tech_pvt->awsAccessKeyId, awsAccessKeyId, 128);
+			strncpy(tech_pvt->awsSecretAccessKey, awsSecretAccessKey, 128);
+			strncpy(tech_pvt->awsSessionToken, awsSessionToken, MAX_SESSION_TOKEN_LEN);
+			strncpy(tech_pvt->region, awsRegion, MAX_REGION);
+		}
+		else if (std::getenv("AWS_ACCESS_KEY_ID") &&
+			std::getenv("AWS_SECRET_ACCESS_KEY") &&
+			std::getenv("AWS_REGION")) {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "Using env vars for aws authentication\n");
+			strncpy(tech_pvt->awsAccessKeyId, std::getenv("AWS_ACCESS_KEY_ID"), 128);
+			strncpy(tech_pvt->awsSecretAccessKey, std::getenv("AWS_SECRET_ACCESS_KEY"), 128);		
+			strncpy(tech_pvt->region, std::getenv("AWS_REGION"), MAX_REGION);
+		}
+		else {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "No channel vars or env vars for aws authentication..will use default profile if found\n");
+		}
+
+		tech_pvt->responseHandler = responseHandler;
+
+		tech_pvt->interim = interim;
+		strncpy(tech_pvt->lang, lang, MAX_LANG);
+		tech_pvt->samples_per_second = sampleRate;
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "sample rate of rtp stream is %d\n", samples_per_second);
+
+    const char* vocabularyName = switch_channel_get_variable(channel, "AWS_VOCABULARY_NAME");
+    const char* vocabularyFilterName = switch_channel_get_variable(channel, "AWS_VOCABULARY_FILTER_NAME");
+    const char* vocabularyFilterMethod = switch_channel_get_variable(channel, "AWS_VOCABULARY_FILTER_METHOD");
+    const char* piiEntityTypes = switch_channel_get_variable(channel, "AWS_PII_ENTITY_TYPES");
+    int shouldIdentifyPII = switch_true(switch_channel_get_variable(channel, "AWS_PII_IDENTIFY_ENTITIES"));
+    const char* languageModelName = switch_channel_get_variable(channel, "AWS_LANGUAGE_MODEL_NAME");
+
+    std::string host, path;
+    TranscribeManager::getSignedWebsocketUrl(
+      host,
+      path,
+      tech_pvt->awsAccessKeyId,
+      tech_pvt->awsSecretAccessKey, 
+      tech_pvt->awsSessionToken, 
+      tech_pvt->region,
+      lang,
+      vocabularyName,
+      vocabularyFilterName,
+      vocabularyFilterMethod,
+      piiEntityTypes,
+      shouldIdentifyPII,
+      languageModelName
+    );
+
+    host = host.substr(0, host.find(':'));
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connecting to host %s, path %s\n", host.c_str(), path.c_str());
+
+    strncpy(tech_pvt->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
+    strncpy(tech_pvt->host, host.c_str(), MAX_WS_URL_LEN);
+    tech_pvt->port = 8443;
+    strncpy(tech_pvt->path, path.c_str(), MAX_PATH_LEN);    
+    tech_pvt->responseHandler = responseHandler;
+    tech_pvt->channels = channels;
+    tech_pvt->id = ++idxCallCount;
+    tech_pvt->buffer_overrun_notified = 0;
+
+    size_t buflen = LWS_PRE + (FRAME_SIZE_8000 * desiredSampling / 8000 * channels * 1000 / RTP_PACKETIZATION_PERIOD * nAudioBufferSecs);
+
+    aws::AudioPipe* ap = new aws::AudioPipe(tech_pvt->sessionId, tech_pvt->bugname, tech_pvt->host, tech_pvt->port, tech_pvt->path, 
+      buflen, read_impl.decoded_bytes_per_packet, eventCallback);
+    if (!ap) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error allocating AudioPipe\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    tech_pvt->pAudioPipe = static_cast<void *>(ap);
+
+		if (switch_mutex_init(&tech_pvt->mutex, SWITCH_MUTEX_NESTED, pool) != SWITCH_STATUS_SUCCESS) {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing mutex\n");
+			status = SWITCH_STATUS_FALSE;
+			goto done; 
+		}
+		if (sampleRate != 8000) {
+			tech_pvt->resampler = speex_resampler_init(1, sampleRate, 16000, SWITCH_RESAMPLE_QUALITY, &err);
+			if (0 != err) {
+				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing resampler: %s.\n", 
+							switch_channel_get_name(channel), speex_resampler_strerror(err));
+				status = SWITCH_STATUS_FALSE;
+				goto done;
+			}
+		}
+
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connecting now\n");
+    ap->connect();
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection in progress\n");
+
+
+		*ppUserData = tech_pvt;
+	
+	done:
+		return status;
+	}
+
+	switch_status_t aws_transcribe_session_stop(switch_core_session_t *session,int channelIsClosing, char* bugname) {
+    switch_channel_t *channel = switch_core_session_get_channel(session);
+    switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME);
+    if (!bug) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "aws_transcribe_session_stop: no bug - websocket conection already closed\n");
+      return SWITCH_STATUS_FALSE;
+    }
+    private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
+    uint32_t id = tech_pvt->id;
+
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) aws_transcribe_session_stop\n", id);
+
+    if (!tech_pvt) return SWITCH_STATUS_FALSE;
+      
+    // close connection and get final responses
+    switch_mutex_lock(tech_pvt->mutex);
+    switch_channel_set_private(channel, bugname, NULL);
+    if (!channelIsClosing) switch_core_media_bug_remove(session, &bug);
+
+    aws::AudioPipe *pAudioPipe = static_cast<aws::AudioPipe *>(tech_pvt->pAudioPipe);
+    if (pAudioPipe) reaper(tech_pvt);
+    destroy_tech_pvt(tech_pvt);
+    switch_mutex_unlock(tech_pvt->mutex);
+    switch_mutex_destroy(tech_pvt->mutex);
+    tech_pvt->mutex = nullptr;
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) aws_transcribe_session_stop\n", id);
+    return SWITCH_STATUS_SUCCESS;
+  }
+	
+	switch_bool_t aws_transcribe_frame(switch_media_bug_t *bug, void* user_data) {
+    switch_core_session_t *session = switch_core_media_bug_get_session(bug);
+    private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
+    size_t inuse = 0;
+    bool dirty = false;
+    char *p = (char *) "{\"msg\": \"buffer overrun\"}";
+
+    if (!tech_pvt) return SWITCH_TRUE;
+    
+    if (switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
+      if (!tech_pvt->pAudioPipe) {
+        switch_mutex_unlock(tech_pvt->mutex);
+        return SWITCH_TRUE;
+      }
+      aws::AudioPipe *pAudioPipe = static_cast<aws::AudioPipe *>(tech_pvt->pAudioPipe);
+      if (pAudioPipe->getLwsState() != aws::AudioPipe::LWS_CLIENT_CONNECTED) {
+        switch_mutex_unlock(tech_pvt->mutex);
+        return SWITCH_TRUE;
+      }
+
+      pAudioPipe->lockAudioBuffer();
+      size_t available = pAudioPipe->binarySpaceAvailable();
+      if (NULL == tech_pvt->resampler) {
+        switch_frame_t frame = { 0 };
+        frame.data = pAudioPipe->binaryWritePtr();
+        frame.buflen = available;
+        while (true) {
+
+          // check if buffer would be overwritten; dump packets if so
+          if (available < pAudioPipe->binaryMinSpace()) {
+            if (!tech_pvt->buffer_overrun_notified) {
+              tech_pvt->buffer_overrun_notified = 1;
+              tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_BUFFER_OVERRUN, NULL, tech_pvt->bugname, 0);
+            }
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "(%u) dropping packets!\n", 
+              tech_pvt->id);
+            pAudioPipe->binaryWritePtrResetToZero();
+
+            frame.data = pAudioPipe->binaryWritePtr();
+            frame.buflen = available = pAudioPipe->binarySpaceAvailable();
+          }
+
+          switch_status_t rv = switch_core_media_bug_read(bug, &frame, SWITCH_TRUE);
+          if (rv != SWITCH_STATUS_SUCCESS) break;
+          if (frame.datalen) {
+            pAudioPipe->binaryWritePtrAdd(frame.datalen);
+            frame.buflen = available = pAudioPipe->binarySpaceAvailable();
+            frame.data = pAudioPipe->binaryWritePtr();
+            dirty = true;
+          }
+        }
+      }
+      else {
+        uint8_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
+        switch_frame_t frame = { 0 };
+        frame.data = data;
+        frame.buflen = SWITCH_RECOMMENDED_BUFFER_SIZE;
+        while (switch_core_media_bug_read(bug, &frame, SWITCH_TRUE) == SWITCH_STATUS_SUCCESS) {
+          if (frame.datalen) {
+            spx_uint32_t out_len = available >> 1;  // space for samples which are 2 bytes
+            spx_uint32_t in_len = frame.samples;
+
+            speex_resampler_process_interleaved_int(tech_pvt->resampler, 
+              (const spx_int16_t *) frame.data, 
+              (spx_uint32_t *) &in_len, 
+              (spx_int16_t *) ((char *) pAudioPipe->binaryWritePtr()),
+              &out_len);
+
+            if (out_len > 0) {
+              // bytes written = num samples * 2 * num channels
+              size_t bytes_written = out_len << tech_pvt->channels;
+              pAudioPipe->binaryWritePtrAdd(bytes_written);
+              available = pAudioPipe->binarySpaceAvailable();
+              dirty = true;
+            }
+            if (available < pAudioPipe->binaryMinSpace()) {
+              if (!tech_pvt->buffer_overrun_notified) {
+                tech_pvt->buffer_overrun_notified = 1;
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "(%u) dropping packets!\n", 
+                  tech_pvt->id);
+                tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_BUFFER_OVERRUN, NULL, tech_pvt->bugname, 0);
+              }
+              break;
+            }
+          }
+        }
+      }
+
+      pAudioPipe->unlockAudioBuffer();
+      switch_mutex_unlock(tech_pvt->mutex);
+    }
+    return SWITCH_TRUE;
+  }
+}

mod_aws_transcribe_ws/aws_transcribe_glue.h

@@ -0,0 +1,11 @@
+#ifndef __AWS_GLUE_H__
+#define __AWS_GLUE_H__
+
+switch_status_t aws_transcribe_init();
+switch_status_t aws_transcribe_cleanup();
+switch_status_t aws_transcribe_session_init(switch_core_session_t *session, responseHandler_t responseHandler, 
+		uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char* bugname, void **ppUserData);
+switch_status_t aws_transcribe_session_stop(switch_core_session_t *session, int channelIsClosing, char* bugname);
+switch_bool_t aws_transcribe_frame(switch_media_bug_t *bug, void* user_data);
+
+#endif

mod_aws_transcribe_ws/mod_aws_transcribe_ws.c

@@ -0,0 +1,212 @@
+/* 
+ *
+ * mod_aws_transcribe.c -- Freeswitch module for using aws streaming transcribe api
+ *
+ */
+#include "mod_aws_transcribe_ws.h"
+#include "aws_transcribe_glue.h"
+
+/* Prototypes */
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_aws_transcribe_ws_shutdown);
+SWITCH_MODULE_LOAD_FUNCTION(mod_aws_transcribe_ws_load);
+
+SWITCH_MODULE_DEFINITION(mod_aws_transcribe_ws, mod_aws_transcribe_ws_load, mod_aws_transcribe_ws_shutdown, NULL);
+
+static switch_status_t do_stop(switch_core_session_t *session, char* bugname);
+
+static void responseHandler(switch_core_session_t* session, 
+	const char* eventName, const char * json, const char* bugname, int finished) {
+	switch_event_t *event;
+	switch_channel_t *channel = switch_core_session_get_channel(session);
+
+	switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, eventName);
+	switch_channel_event_set_data(channel, event);
+	switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-vendor", "deepgram");
+	switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-session-finished", finished ? "true" : "false");
+	if (finished) {
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "responseHandler returning event %s, from finished recognition session\n", eventName);
+	}
+	if (json) switch_event_add_body(event, "%s", json);
+	if (bugname) switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "media-bugname", bugname);
+	switch_event_fire(&event);
+}
+
+
+static switch_bool_t capture_callback(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type)
+{
+	switch_core_session_t *session = switch_core_media_bug_get_session(bug);
+
+	switch (type) {
+	case SWITCH_ABC_TYPE_INIT:
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Got SWITCH_ABC_TYPE_INIT.\n");
+		break;
+
+	case SWITCH_ABC_TYPE_CLOSE:
+		{
+			private_t *tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Got SWITCH_ABC_TYPE_CLOSE.\n");
+			aws_transcribe_session_stop(session, 1, tech_pvt->bugname);
+			//switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Finished SWITCH_ABC_TYPE_CLOSE.\n");
+		}
+		break;
+	
+	case SWITCH_ABC_TYPE_READ:
+
+		return aws_transcribe_frame(bug, user_data);
+		break;
+
+	case SWITCH_ABC_TYPE_WRITE:
+	default:
+		break;
+	}
+
+	return SWITCH_TRUE;
+}
+
+static switch_status_t start_capture(switch_core_session_t *session, switch_media_bug_flag_t flags, 
+  char* lang, int interim, char* bugname)
+{
+	switch_channel_t *channel = switch_core_session_get_channel(session);
+	switch_media_bug_t *bug;
+	switch_status_t status;
+	switch_codec_implementation_t read_impl = { 0 };
+	void *pUserData;
+	uint32_t samples_per_second;
+
+	if (switch_channel_get_private(channel, bugname)) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing bug from previous transcribe\n");
+		do_stop(session, bugname);
+	}
+
+	switch_core_session_get_read_impl(session, &read_impl);
+
+	if (switch_channel_pre_answer(channel) != SWITCH_STATUS_SUCCESS) {
+		return SWITCH_STATUS_FALSE;
+	}
+
+	samples_per_second = !strcasecmp(read_impl.iananame, "g722") ? read_impl.actual_samples_per_second : read_impl.samples_per_second;
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, " initializing aws speech session.\n");
+	if (SWITCH_STATUS_FALSE == aws_transcribe_session_init(session, responseHandler, samples_per_second, 
+    flags & SMBF_STEREO ? 2 : 1, lang, interim, bugname, &pUserData)) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing aws speech session.\n");
+		return SWITCH_STATUS_FALSE;
+	}
+	if ((status = switch_core_media_bug_add(session, bugname, NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
+		return status;
+	}
+  switch_channel_set_private(channel, bugname, bug);
+	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for aws transcribe\n");
+
+	return SWITCH_STATUS_SUCCESS;
+}
+
+static switch_status_t do_stop(switch_core_session_t *session, char* bugname)
+{
+	switch_status_t status = SWITCH_STATUS_SUCCESS;
+
+	switch_channel_t *channel = switch_core_session_get_channel(session);
+	switch_media_bug_t *bug = switch_channel_get_private(channel, bugname);
+
+	if (bug) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Received user command command to stop transcribe on %s.\n", bugname);
+		status = aws_transcribe_session_stop(session, 0, bugname);
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "stopped transcribe.\n");
+	}
+
+	return status;
+}
+
+#define TRANSCRIBE_API_SYNTAX "<uuid> [start|stop] lang-code [interim] [stereo|mono] [bugname]"
+SWITCH_STANDARD_API(aws_transcribe_function)
+{
+	char *mycmd = NULL, *argv[6] = { 0 };
+	int argc = 0;
+	switch_status_t status = SWITCH_STATUS_FALSE;
+	switch_media_bug_flag_t flags = SMBF_READ_STREAM /* | SMBF_WRITE_STREAM | SMBF_READ_PING */;
+
+	if (!zstr(cmd) && (mycmd = strdup(cmd))) {
+		argc = switch_separate_string(mycmd, ' ', argv, (sizeof(argv) / sizeof(argv[0])));
+	}
+
+	if (zstr(cmd) || 
+      (!strcasecmp(argv[1], "stop") && argc < 2) ||
+      (!strcasecmp(argv[1], "start") && argc < 3) ||
+      zstr(argv[0])) {
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error with command %s %s %s.\n", cmd, argv[0], argv[1]);
+		stream->write_function(stream, "-USAGE: %s\n", TRANSCRIBE_API_SYNTAX);
+		goto done;
+	} else {
+		switch_core_session_t *lsession = NULL;
+
+		if ((lsession = switch_core_session_locate(argv[0]))) {
+			if (!strcasecmp(argv[1], "stop")) {
+				char *bugname = argc > 2 ? argv[2] : MY_BUG_NAME;
+    		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "stop transcribing\n");
+				status = do_stop(lsession, bugname);
+			} else if (!strcasecmp(argv[1], "start")) {
+        char* lang = argv[2];
+        int interim = argc > 3 && !strcmp(argv[3], "interim");
+				char *bugname = argc > 5 ? argv[5] : MY_BUG_NAME;
+				if (argc > 4 && !strcmp(argv[4], "stereo")) {
+          flags |= SMBF_WRITE_STREAM ;
+          flags |= SMBF_STEREO;
+				}
+    		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "start transcribing %s %s %s\n", lang, interim ? "interim": "complete", bugname);
+				status = start_capture(lsession, flags, lang, interim, bugname);
+			}
+			switch_core_session_rwunlock(lsession);
+		}
+	}
+
+	if (status == SWITCH_STATUS_SUCCESS) {
+		stream->write_function(stream, "+OK Success\n");
+	} else {
+		stream->write_function(stream, "-ERR Operation Failed\n");
+	}
+
+  done:
+
+	switch_safe_free(mycmd);
+	return SWITCH_STATUS_SUCCESS;
+}
+
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_aws_transcribe_ws_load)
+{
+	switch_api_interface_t *api_interface;
+
+	/* create/register custom event message type */
+	if (switch_event_reserve_subclass(TRANSCRIBE_EVENT_RESULTS) != SWITCH_STATUS_SUCCESS) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't register subclass %s!\n", TRANSCRIBE_EVENT_RESULTS);
+		return SWITCH_STATUS_TERM;
+	}
+
+	/* connect my internal structure to the blank pointer passed to me */
+	*module_interface = switch_loadable_module_create_module_interface(pool, modname);
+
+	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AWS Speech Transcription API loading..\n");
+
+  if (SWITCH_STATUS_FALSE == aws_transcribe_init()) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, "Failed initializing aws speech interface\n");
+	}
+
+	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AWS Speech Transcription API successfully loaded\n");
+
+	SWITCH_ADD_API(api_interface, "uuid_aws_transcribe", "AWS Speech Transcription API", aws_transcribe_function, TRANSCRIBE_API_SYNTAX);
+	switch_console_set_complete("add uuid_aws_transcribe start lang-code [interim|final] [stereo|mono]");
+	switch_console_set_complete("add uuid_aws_transcribe stop ");
+
+	/* indicate that the module should continue to be loaded */
+	return SWITCH_STATUS_SUCCESS;
+}
+
+/*
+  Called when the system shuts down
+  Macro expands to: switch_status_t mod_aws_transcribe_ws_shutdown() */
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_aws_transcribe_ws_shutdown)
+{
+	aws_transcribe_cleanup();
+	switch_event_free_subclass(TRANSCRIBE_EVENT_RESULTS);
+	return SWITCH_STATUS_SUCCESS;
+}

mod_aws_transcribe_ws/mod_aws_transcribe_ws.h

@@ -0,0 +1,60 @@
+#ifndef __MOD_AWS_TRANSCRIBE_WS_H__
+#define __MOD_AWS_TRANSCRIBE_WS_H__
+
+#include <switch.h>
+#include <speex/speex_resampler.h>
+
+#include <unistd.h>
+
+#define MY_BUG_NAME "aws_transcribe_ws"
+#define MAX_BUG_LEN (64)
+#define MAX_SESSION_ID (256)
+#define TRANSCRIBE_EVENT_RESULTS "aws_transcribe::transcription"
+#define TRANSCRIBE_EVENT_END_OF_TRANSCRIPT "aws_transcribe::end_of_transcript"
+#define TRANSCRIBE_EVENT_NO_AUDIO_DETECTED "aws_transcribe::no_audio_detected"
+#define TRANSCRIBE_EVENT_MAX_DURATION_EXCEEDED "aws_transcribe::max_duration_exceeded"
+#define TRANSCRIBE_EVENT_VAD_DETECTED "aws_transcribe::vad_detected"
+#define TRANSCRIBE_EVENT_CONNECT_SUCCESS "aws::connect"
+#define TRANSCRIBE_EVENT_CONNECT_FAIL    "aws::connect_failed"
+#define TRANSCRIBE_EVENT_DISCONNECT      "aws::disconnect"
+#define TRANSCRIBE_EVENT_BUFFER_OVERRUN  "aws::buffer_overrun"
+#define TRANSCRIBE_EVENT_ERROR      "jambonz_transcribe::error"
+
+#define MAX_LANG (12)
+#define MAX_REGION (32)
+#define MAX_WS_URL_LEN (512)
+#define MAX_PATH_LEN (4096)
+#define MAX_SESSION_TOKEN_LEN (4096)
+
+/* per-channel data */
+typedef void (*responseHandler_t)(switch_core_session_t* session, const char* eventName, const char* json, const char* bugname, int finished);
+
+struct private_data {
+	switch_mutex_t *mutex;
+	char sessionId[MAX_SESSION_ID+1];
+  char awsAccessKeyId[128];
+  char awsSecretAccessKey[128];
+  char awsSessionToken[MAX_SESSION_TOKEN_LEN+1];
+  SpeexResamplerState *resampler;
+	responseHandler_t responseHandler;
+	int interim;
+	char lang[MAX_LANG+1];
+	char region[MAX_REGION+1];
+	switch_vad_t * vad;
+	uint32_t samples_per_second;
+  void *pAudioPipe;
+  int ws_state;
+  char host[MAX_WS_URL_LEN+1];
+  unsigned int port;
+  char path[MAX_PATH_LEN+1];
+  char bugname[MAX_BUG_LEN+1];
+  int sampling;
+  int  channels;
+  unsigned int id;
+  int buffer_overrun_notified:1;
+  int is_finished:1;
+};
+
+typedef struct private_data private_t;
+
+#endif

mod_aws_transcribe_ws/simple_buffer.h

@@ -0,0 +1,51 @@
+/**
+ * (very) simple and limited circular buffer, 
+ * supporting only the use case of doing all of the adds
+ * and then subsquently retrieves.
+ * 
+ */
+class SimpleBuffer {
+  public:
+    SimpleBuffer(uint32_t chunkSize, uint32_t numChunks) : numItems(0),
+    m_numChunks(numChunks), m_chunkSize(chunkSize) {
+      m_pData = new char[chunkSize * numChunks];
+      m_pNextWrite = m_pData;
+    }
+    ~SimpleBuffer() {
+      delete [] m_pData;
+    }
+
+    void add(void *data, uint32_t datalen) {
+      if (datalen % m_chunkSize != 0) return;
+      int numChunks = datalen / m_chunkSize;
+      for (int i = 0; i < numChunks; i++) {
+        memcpy(m_pNextWrite, data, m_chunkSize);
+        data = static_cast<char*>(data) + m_chunkSize;
+        if (numItems < m_numChunks) numItems++;
+
+        uint32_t offset = (m_pNextWrite - m_pData) / m_chunkSize;
+        if (offset >= m_numChunks - 1) m_pNextWrite = m_pData;
+        else m_pNextWrite += m_chunkSize;
+      }
+    }
+
+    char* getNextChunk() {
+      if (numItems--) {
+        char *p = m_pNextWrite;
+        uint32_t offset = (m_pNextWrite - m_pData) / m_chunkSize;
+        if (offset >= m_numChunks - 1) m_pNextWrite = m_pData;
+        else m_pNextWrite += m_chunkSize;
+        return p;
+      }
+      return nullptr;
+    }
+
+    uint32_t getNumItems() { return numItems;}
+
+  private:
+    char *m_pData;
+    uint32_t numItems;
+    uint32_t m_chunkSize;
+    uint32_t m_numChunks;
+    char* m_pNextWrite;
+};

mod_aws_transcribe_ws/transcribe_manager.cpp

@@ -0,0 +1,324 @@
+#include "transcribe_manager.hpp"
+#include "crc.h"
+
+#include <switch.h>
+
+#include <openssl/sha.h>
+#include <openssl/hmac.h>
+#include <iomanip>
+#include <regex>
+#include <iostream>
+#include <cstring>
+#include <netinet/in.h>
+
+using namespace std;
+
+namespace {
+  std::string uri_encode(const std::string &value) {
+    std::string encoded;
+    char hex[4];
+    for (char c : value) {
+        if (isalnum(c) || c == '-' || c == '_' || c == '.' || c == '~') {
+            encoded += c;
+        } else {
+            sprintf(hex, "%%%02X", c);
+            encoded.append(hex);
+        }
+    }
+    return encoded;
+  }
+}
+// see
+// https://docs.aws.amazon.com/transcribe/latest/dg/websocket.html#websocket-url
+// https://docs.aws.amazon.com/transcribe/latest/dg/event-stream.html
+
+void TranscribeManager::getSignedWebsocketUrl(string& host, string& path, const string& accessKey,
+        const string& secretKey, const string& securityToken, const string& region, const std::string& lang, 
+        const char* vocabularyName, const char* vocabularyFilterName, const char* vocabularyFilterMethod,
+        const char* piiEntities, int shouldIdentifyPiiEntities, const char* languageModelName) {
+    string method = "GET";
+    string service = "transcribe";
+    string endpoint = "wss://transcribestreaming." + region + ".amazonaws.com";
+    host = "transcribestreaming." + region + ".amazonaws.com";
+
+    time_t now = time(0);
+    tm *gmtm = gmtime(&now);
+
+    char amzDate[21];
+    snprintf (amzDate, 21, "%04d%02d%02dT%02d%02d%02dZ",
+            1900 + gmtm->tm_year, 1 + gmtm->tm_mon, gmtm->tm_mday,
+            gmtm->tm_hour, gmtm->tm_min, gmtm->tm_sec);
+
+    char datestamp[9];
+    snprintf (datestamp, 9, "%04d%02d%02d", 1900 + gmtm->tm_year, 1 + gmtm->tm_mon, gmtm->tm_mday);
+
+    string canonical_uri = "/stream-transcription-websocket";
+    string canonical_headers = "host:" + host + "\n";
+    string signed_headers = "host";
+    string algorithm = "AWS4-HMAC-SHA256";
+    string credential_scope = string(datestamp) + "%2F" + region + "%2F" + service + "%2F" + "aws4_request";
+
+    // N.B.: The order of all of these query args are important!
+    // Otherwise, the signature will be invalid.
+    string canonical_querystring  = "X-Amz-Algorithm=" + algorithm;
+    canonical_querystring += "&X-Amz-Credential=" + accessKey + "%2F" + credential_scope;
+    canonical_querystring += "&X-Amz-Date=" + string(amzDate);
+    canonical_querystring += "&X-Amz-Expires=300";
+    canonical_querystring += "&X-Amz-Security-Token=" + uri_encode(securityToken);
+    canonical_querystring += "&X-Amz-SignedHeaders=" + signed_headers;
+
+    if (piiEntities && shouldIdentifyPiiEntities) {
+      canonical_querystring += "&content-redaction-type=PII";
+    }
+    
+    canonical_querystring += "&language-code=" + lang;
+    
+    if (languageModelName) {
+      std::string str(languageModelName);
+      canonical_querystring += "&language-model-name=" + uri_encode(str);
+    }
+
+    canonical_querystring += "&media-encoding=pcm";
+
+    if (piiEntities) {
+      std::string str(piiEntities);
+      canonical_querystring += "&pii-entitytypes=" + uri_encode(str);
+    }
+
+    canonical_querystring += "&sample-rate=8000";
+
+    // custom vocabulary and filter
+    if (vocabularyFilterMethod) {
+      std::string str(vocabularyFilterMethod);
+      canonical_querystring += "&vocabulary-filter-method=" + str;
+    }
+    if (vocabularyFilterName) {
+      std::string str(vocabularyFilterName);
+      canonical_querystring += "&vocabulary-filter-name=" + str;
+    }
+    if (vocabularyName) {
+      std::string str(vocabularyName);
+      canonical_querystring += "&vocabulary-name=" + str;
+    }
+
+    string payload_hash = getSha256("");
+
+    string canonical_request = method + '\n'
+       + canonical_uri + '\n'
+       + canonical_querystring + '\n'
+       + canonical_headers + '\n'
+       + signed_headers + '\n'
+       + payload_hash;
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"TranscribeManager::getSignedWebsocketUrl canonical_request: %s\n", canonical_request.c_str());
+    string string_to_sign = algorithm + "\n"
+       + amzDate + "\n"
+       + regex_replace(credential_scope, regex("%2F"), "/") + "\n"
+       + getSha256(canonical_request);
+
+    unsigned char signing_key[SHA256_DIGEST_LENGTH];
+    getSignatureKey(signing_key, secretKey, datestamp, region, service);
+
+    unsigned char signatureBinary[SHA256_DIGEST_LENGTH];
+    getHMAC(signatureBinary, signing_key, SHA256_DIGEST_LENGTH, string_to_sign);
+    string signature = toHex(signatureBinary);
+
+    canonical_querystring += "&X-Amz-Signature=" + signature;
+    string request_url = endpoint + canonical_uri + "?" + canonical_querystring;
+    path = canonical_uri + "?" + canonical_querystring;
+
+    return;
+}
+
+string TranscribeManager::getSha256(string str) {
+    SHA256_CTX ctx;
+    SHA256_Init(&ctx);
+    SHA256_Update(&ctx, str.c_str(), str.length());
+    unsigned char hash[SHA256_DIGEST_LENGTH] = { 0 };
+    SHA256_Final(hash, &ctx);
+
+    ostringstream os;
+    os << hex << setfill('0');
+    for (int i = 0; i < SHA256_DIGEST_LENGTH; ++i) {
+        os << setw(2) << static_cast<unsigned int>(hash[i]);
+    }
+
+    return os.str();
+}
+
+void TranscribeManager::getSignatureKey(unsigned char *signatureKey, const string& secretKey,
+            const string& datestamp, const string& region, const string& service) {
+    string key = string("AWS4") + secretKey;
+    unsigned char kDate[SHA256_DIGEST_LENGTH];
+    unsigned char kRegion[SHA256_DIGEST_LENGTH];
+    unsigned char kService[SHA256_DIGEST_LENGTH];
+    unsigned char kSigning[SHA256_DIGEST_LENGTH];
+    getHMAC(kDate, (unsigned char *)key.c_str(), key.length(), datestamp);
+    getHMAC(kRegion, kDate, SHA256_DIGEST_LENGTH, region);
+    getHMAC(kService, kRegion, SHA256_DIGEST_LENGTH, service);
+    getHMAC(kSigning, kService, SHA256_DIGEST_LENGTH, "aws4_request");
+
+    memcpy(signatureKey, kSigning, SHA256_DIGEST_LENGTH);
+}
+
+void TranscribeManager::getHMAC(unsigned char *hmac, unsigned char *key, int keyLen, const string& str) {
+    unsigned char *data = (unsigned char*)str.c_str();
+    unsigned char *result = HMAC(EVP_sha256(), key, keyLen, data, strlen((char *)data), NULL, NULL);
+    memcpy(hmac, result, SHA256_DIGEST_LENGTH);
+}
+
+string TranscribeManager::toHex(unsigned char *hmac) {
+    ostringstream os;
+    os << hex << setfill('0');
+    for (int i = 0; i < SHA256_DIGEST_LENGTH; ++i) {
+        os << setw(2) << static_cast<unsigned int>(hmac[i]);
+    }
+
+    return os.str();
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////
+
+bool TranscribeManager::parseResponse(const string& response, string& payload, bool& isError, bool verbose) {
+    const char* buffer = response.c_str();
+
+    uint32_t totalLen;
+    memcpy(&totalLen, &buffer[0], sizeof(uint32_t));
+    totalLen = ntohl(totalLen);
+
+    uint32_t headerLen;
+    memcpy(&headerLen, &buffer[4], sizeof(uint32_t));
+    headerLen = ntohl(headerLen);
+
+    if (!verifyCRC(buffer, totalLen)) {
+        return false;
+    }
+
+    buffer += 12; // bytes 0 - 11 are prelude
+
+    const int numberOfHeaders = 3;
+    for (int i = 0; i < numberOfHeaders; i++) {
+        parseHeader(&buffer, isError, verbose);
+    }
+
+    payload = string(buffer, totalLen - headerLen - 4*4);
+
+    return true;
+}
+
+bool TranscribeManager::verifyCRC(const char* buffer, const uint32_t totalLength) {
+    uint32_t preludeCRC;
+    memcpy(&preludeCRC, &buffer[8], 4);
+    preludeCRC = ntohl(preludeCRC);
+
+    uint32_t calculatedPreludeCRC = CRC::Calculate(&buffer[0], 8, CRC::CRC_32());
+    if (calculatedPreludeCRC != preludeCRC) {
+        cout << "Prelude CRC didn't match!" << endl;
+        return false;
+    }
+
+    uint32_t messageCRC;
+    memcpy(&messageCRC, &buffer[totalLength - 4], 4);
+    messageCRC = ntohl(messageCRC);
+
+    uint32_t calculatedMessageCRC = CRC::Calculate(buffer, totalLength - 4, CRC::CRC_32());
+
+    if (calculatedMessageCRC != messageCRC) {
+        cout << "Message CRC didn't match!" << endl;
+        return false;
+    }
+
+    return true;
+}
+
+void TranscribeManager::parseHeader(const char** buffer, bool& isError, bool verbose) {
+    uint8_t headerNameLen;
+    memcpy(&headerNameLen, *buffer, sizeof(uint8_t));
+    (*buffer)++;
+
+    string headerName(*buffer, headerNameLen);
+    *buffer += headerNameLen;
+
+    uint8_t headerType;
+    memcpy(&headerType, *buffer, sizeof(uint8_t));
+    (*buffer)++;
+
+    uint16_t headerValLen;
+    memcpy(&headerValLen, *buffer, sizeof(uint16_t));
+    headerValLen = ntohs(headerValLen);
+    *buffer += 2;
+
+    string headerVal(*buffer, headerValLen);
+    *buffer += headerValLen;
+
+    if (headerVal == "exception") {
+        isError = true;
+    }
+    if (verbose) {
+        cout << headerName << "(" << (int)headerType << "): " << headerVal << endl;
+    }
+}
+
+///////////////////////////////////////////////////////////////////////////////////////////
+
+bool TranscribeManager::makeRequest(string& request, const vector<uint8_t>& data) {
+    char preludeAndCrcBuffer[4*3];
+    char headerBuffer[88];
+    char messageCrcBuffer[4];
+
+    // prelude
+    uint32_t totalLen = sizeof(preludeAndCrcBuffer) + sizeof(headerBuffer) + data.size() + sizeof(messageCrcBuffer);
+    uint32_t headerLen = sizeof(headerBuffer);
+    totalLen = htonl(totalLen);
+    headerLen = htonl(headerLen);
+
+    memcpy(&preludeAndCrcBuffer[0], &totalLen, sizeof(uint32_t));
+    memcpy(&preludeAndCrcBuffer[4], &headerLen, sizeof(uint32_t));
+
+    uint32_t preludeCRC = CRC::Calculate(&preludeAndCrcBuffer[0], 8, CRC::CRC_32());
+    preludeCRC = htonl(preludeCRC);
+    memcpy(&preludeAndCrcBuffer[8], &preludeCRC, sizeof(uint32_t));
+
+    // header
+    char* buffer = headerBuffer;
+    writeHeader(&buffer, ":content-type", "application/octet-stream");
+    writeHeader(&buffer, ":event-type", "AudioEvent");
+    writeHeader(&buffer, ":message-type", "event");
+
+    // write everything to response string except for the message CRC
+    request.append(preludeAndCrcBuffer, sizeof(preludeAndCrcBuffer));
+    request.append(headerBuffer, sizeof(headerBuffer));
+    request.append(data.begin(), data.end());
+
+    // message CRC
+    uint32_t messageCRC = CRC::Calculate(request.c_str(), request.length(), CRC::CRC_32());
+    messageCRC = htonl(messageCRC);
+    memcpy(messageCrcBuffer, &messageCRC, sizeof(uint32_t));
+
+    // write message CRC to response string
+    request.append(messageCrcBuffer, sizeof(messageCrcBuffer));
+
+    return true;
+}
+
+void TranscribeManager::writeHeader(char** buffer, const char* key, const char* val) {
+    uint8_t keyLen = strlen(key);
+    uint16_t valueLen = strlen(val);
+
+    memcpy(*buffer, &keyLen, sizeof(uint8_t));
+    (*buffer)++;
+
+    memcpy(*buffer, key, keyLen);
+    (*buffer) += keyLen;
+
+    uint8_t valueType = 7;
+    memcpy(*buffer, &valueType, sizeof(uint8_t));
+    (*buffer)++;
+
+    uint16_t valLen = htons(valueLen);
+    memcpy(*buffer, &valLen, sizeof(uint16_t));
+    (*buffer) += 2;
+
+    memcpy(*buffer, val, valueLen);
+    (*buffer) += valueLen;
+}

mod_aws_transcribe_ws/transcribe_manager.hpp

@@ -0,0 +1,50 @@
+#ifndef TRANSCRIBEMANAGER_HPP_
+#define TRANSCRIBEMANAGER_HPP_
+
+#include <string>
+#include <vector>
+
+/** Usage
+ #include "transcribe_manager.hpp"
+
+ // get signed URL
+ const string url = TranscribeManager::getSignedWebsocketUrl(accessKey_, secretKey_, region_);
+
+ // connect to the url using a socket library (e.g. https://github.com/machinezone/IXWebSocket)
+
+ // build request string
+ string request;
+ TranscribeManager::makeRequest(request, audioData); // audioData is a const vector<uint8_t>
+
+ // send request to socket
+ * 
+ */
+
+using namespace std;
+
+class TranscribeManager {
+public:
+    static void getSignedWebsocketUrl(string& host, string& path,
+            const std::string& accessKey, const std::string& secretKey, const std::string& securityToken, 
+            const std::string& region, const std::string& lang, const char* vocabularyName,
+            const char* vocabularyFilterName, const char* vocabularyFilterMethod,
+            const char* piiEntities, int shouldIdentifyPiiEntities, const char* languageModelName);
+
+    static bool parseResponse(const std::string& response, std::string& payload, bool& isError, bool verbose = false);
+
+    static bool makeRequest(std::string& request, const std::vector<uint8_t>& data);
+    static void writeHeader(char** buffer, const char* key, const char* val);
+
+private:
+    static std::string getSha256(std::string str);
+    static void getSignatureKey(unsigned char *signatureKey, const std::string& secretKey,
+            const std::string& datestamp, const std::string& region, const std::string& service);
+    static void getHMAC(unsigned char *hmac, unsigned char *key, int keyLen, const std::string& str);
+    static std::string toHex(unsigned char *hmac);
+
+    static bool verifyCRC(const char* buffer, const uint32_t totalLength);
+    static void parseHeader(const char** buffer, bool& isError, bool verbose = false);
+
+};
+
+#endif /* TRANSCRIBEMANAGER_HPP_ */

mod_azure_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_azure_transcribe/azure_transcribe_glue.cpp

@@ -32,11 +32,12 @@ static const char* proxyIP = std::getenv("JAMBONES_HTTP_PROXY_IP");
 static const char* proxyPort = std::getenv("JAMBONES_HTTP_PROXY_PORT");
 static const char* proxyUsername = std::getenv("JAMBONES_HTTP_PROXY_USERNAME");
 static const char* proxyPassword = std::getenv("JAMBONES_HTTP_PROXY_PASSWORD");
+static const bool use_single_connection = switch_true(std::getenv("AZURE_SPEECH_USE_SINGLE_CONNECTION"));
 
 class GStreamer {
 public:
 	GStreamer(
-    const char *sessionId,
+		const char *sessionId,
 		const char *bugname,
 		u_int16_t channels,
     char *lang, 
@@ -135,6 +136,19 @@ public:
       properties.SetProperty(PropertyId::Speech_SegmentationSilenceTimeoutMs, segmentationInterval);
     }
 
+		//https://learn.microsoft.com/en-us/azure/ai-services/speech-service/language-identification?tabs=once&pivots=programming-language-cpp#at-start-and-continuous-language-identification
+		const char* languageIdMode = switch_channel_get_variable(channel, "AZURE_LANGUAGE_ID_MODE");
+		if (languageIdMode) {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(psession), SWITCH_LOG_DEBUG, "setting SpeechServiceConnection_LanguageIdMode to %s \n", languageIdMode);
+			properties.SetProperty(PropertyId::SpeechServiceConnection_LanguageIdMode, languageIdMode);
+		}
+		//https://learn.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/propertyid?view=azure-node-latest
+		//PropertyId::SpeechServiceResponse_PostProcessingOption
+ 		const char* postProcessingOption = switch_channel_get_variable(channel, "AZURE_POST_PROCESSING_OPTION");
+		if (postProcessingOption) {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(psession), SWITCH_LOG_DEBUG, "setting SpeechServiceResponse_PostProcessingOption to %s \n", postProcessingOption);
+			properties.SetProperty(PropertyId::SpeechServiceResponse_PostProcessingOption, postProcessingOption);
+		}
 		// recognition mode - readonly according to Azure docs: 
 		// https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/propertyid?view=azure-node-latest
 		/*
@@ -239,6 +253,9 @@ public:
 		m_recognizer->Recognized += onRecognitionEvent;
 		m_recognizer->Canceled += onCanceled;
 
+		// Store the final configuration string
+    m_configuration_string = createConfigurationStr(channels, lang, interim, samples_per_second, region, subscriptionKey, psession);
+
 		switch_core_session_rwunlock(psession);
 	}
 
@@ -246,6 +263,10 @@ public:
 		//switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer::~GStreamer %p\n", this);		
 	}
 
+	const char* configuration() {
+		return m_configuration_string.c_str();
+	}
+
 	void connect() {
 		if (m_connecting) return;
 		m_connecting = true;
@@ -311,13 +332,33 @@ public:
     return m_connecting;
   }
 
+	bool hasConfigurationChanged(
+		u_int16_t channels,
+		char *lang, 
+		int interim,
+		uint32_t samples_per_second,
+		const char* region, 
+		const char* subscriptionKey) {
+		switch_core_session_t* psession = switch_core_session_locate(m_sessionId.c_str());
+		if (!psession) throw std::invalid_argument( "session id no longer active" );
+
+		std::string newConfiguration = createConfigurationStr(channels, lang, interim, samples_per_second, region, subscriptionKey, psession);
+
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(psession), SWITCH_LOG_DEBUG,
+			"hasConfigurationChanged: old configurattion: %s, new configuration: %s\n", configuration(),  newConfiguration.c_str());
+
+		switch_core_session_rwunlock(psession);
+
+		return strcmp(newConfiguration.c_str(), configuration());
+	}
+
 private:
 	std::string m_sessionId;
 	std::string m_bugname;
 	std::string  m_region;
 	std::shared_ptr<SpeechRecognizer> m_recognizer;
 	std::shared_ptr<PushAudioInputStream> m_pushStream;
-
+  std::string m_configuration_string;
 	responseHandler_t m_responseHandler;
 	bool m_interim;
 	bool m_finished;
@@ -325,6 +366,64 @@ private:
 	bool m_connecting;
 	bool m_stopped;
 	SimpleBuffer m_audioBuffer;
+
+	std::string createConfigurationStr(
+		u_int16_t channels,
+		char *lang, 
+		int interim,
+		uint32_t samples_per_second,
+		const char* region, 
+		const char* subscriptionKey,
+		switch_core_session_t* psession
+	) {
+		switch_channel_t *channel = switch_core_session_get_channel(psession);
+		std::ostringstream configuration_stream;
+		configuration_stream << 
+			channels << ";" <<
+			lang << ";" <<
+			interim << ";" <<
+			samples_per_second << ";" <<
+			region << ";" <<
+			subscriptionKey << ";";
+
+		const char* endpoint = switch_channel_get_variable(channel, "AZURE_SERVICE_ENDPOINT");
+		const char* endpointId = switch_channel_get_variable(channel, "AZURE_SERVICE_ENDPOINT_ID");
+		configuration_stream << (endpoint ? endpoint : "") << ";"
+			<< (endpointId ? endpointId : "") << ";";
+		if (switch_true(switch_channel_get_variable(channel, "AZURE_USE_OUTPUT_FORMAT_DETAILED"))) {
+			configuration_stream << "output_format_detailed;";
+		}
+		if (switch_true(switch_channel_get_variable(channel, "AZURE_AUDIO_LOGGING"))) {
+			configuration_stream << "audio_logging;";
+		}
+		configuration_stream << (proxyIP ? proxyIP : "") << ";"
+			<< (proxyPort ? proxyPort : "") << ";"
+			<< (proxyUsername ? proxyUsername : "") << ";"
+			<< (proxyPassword ? proxyPassword : "") << ";";
+		const char* var;
+		if (var = switch_channel_get_variable(channel, "AZURE_SPEECH_ALTERNATIVE_LANGUAGE_CODES")) {
+			configuration_stream << var << ";";
+		}
+		if (var = switch_channel_get_variable(channel, "AZURE_PROFANITY_OPTION")) {
+			configuration_stream << var << ";";
+		}
+		if (var = switch_channel_get_variable(channel, "AZURE_REQUEST_SNR")) {
+			configuration_stream << var << ";";
+		}
+		if (var = switch_channel_get_variable(channel, "AZURE_INITIAL_SPEECH_TIMEOUT_MS")) {
+			configuration_stream << var << ";";
+		}
+		if (var = switch_channel_get_variable(channel, "AZURE_SPEECH_SEGMENTATION_SILENCE_TIMEOUT_MS")) {
+			configuration_stream << var << ";";
+		}
+		if (var = switch_channel_get_variable(channel, "AZURE_LANGUAGE_ID_MODE")) {
+			configuration_stream << var << ";";
+		}
+		if (var = switch_channel_get_variable(channel, "AZURE_SPEECH_HINTS")) {
+			configuration_stream << var << ";";
+		}
+		return configuration_stream.str();
+	}
 };
 
 static void reaper(struct cap_cb *cb) {
@@ -381,16 +480,33 @@ extern "C" {
 		GStreamer *streamer = NULL;
 		switch_status_t status = SWITCH_STATUS_SUCCESS;
 		switch_channel_t *channel = switch_core_session_get_channel(session);
+		switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, bugname);
+		const char* subscriptionKey = switch_channel_get_variable(channel, "AZURE_SUBSCRIPTION_KEY");
+		const char* region = switch_channel_get_variable(channel, "AZURE_REGION");
+		const char* sessionId = switch_core_session_get_uuid(session);
+		auto read_codec = switch_core_session_get_read_codec(session);
+		uint32_t sampleRate = read_codec->implementation->actual_samples_per_second;
+		if (bug) {
+			struct cap_cb* existing_cb = (struct cap_cb*) switch_core_media_bug_get_user_data(bug);
+			GStreamer* existing_streamer = (GStreamer*) existing_cb->streamer;
+			existing_cb->is_keep_alive = 0;
+			if (!existing_streamer->hasConfigurationChanged(channels, lang, interim, sampleRate, region, subscriptionKey)) {
+				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "Reuse active azure connection.\n");
+				return SWITCH_STATUS_SUCCESS;
+			}
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "Azure configuration is changed, destroy old and create new azure connection\n");
+			reaper(existing_cb);
+			streamer =  new GStreamer(sessionId, bugname, channels, lang, interim, sampleRate, region, subscriptionKey, responseHandler);
+			if (!existing_cb->vad) streamer->connect();
+			existing_cb->streamer = streamer;
+			*ppUserData = existing_cb;
+			return SWITCH_STATUS_SUCCESS;
+		}
 		int err;
 		switch_threadattr_t *thd_attr = NULL;
 		switch_memory_pool_t *pool = switch_core_session_get_pool(session);
-		auto read_codec = switch_core_session_get_read_codec(session);
-		uint32_t sampleRate = read_codec->implementation->actual_samples_per_second;
-		const char* sessionId = switch_core_session_get_uuid(session);
 		struct cap_cb* cb = (struct cap_cb *) switch_core_session_alloc(session, sizeof(*cb));
 		memset(cb, sizeof(cb), 0);
-		const char* subscriptionKey = switch_channel_get_variable(channel, "AZURE_SUBSCRIPTION_KEY");
-		const char* region = switch_channel_get_variable(channel, "AZURE_REGION");
 		cb->channels = channels;
 		strncpy(cb->sessionId, sessionId, MAX_SESSION_ID);
 		strncpy(cb->bugname, bugname, MAX_BUG_LEN);
@@ -411,15 +527,27 @@ extern "C" {
 
 		cb->responseHandler = responseHandler;
 
+		cb->interim = interim;
+		strncpy(cb->lang, lang, MAX_LANG);
+
+		try {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "%s: initializing gstreamer with %s\n", 
+					switch_channel_get_name(channel), bugname);
+			streamer = new GStreamer(sessionId, bugname, channels, lang, interim, sampleRate, cb->region, subscriptionKey, responseHandler);
+			cb->streamer = streamer;
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "azure_transcribe_session_init: config: %s\n", streamer->configuration());
+		} catch (std::exception& e) {
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing gstreamer: %s.\n", 
+				switch_channel_get_name(channel), e.what());
+			return SWITCH_STATUS_FALSE;
+		}
+
 		if (switch_mutex_init(&cb->mutex, SWITCH_MUTEX_NESTED, pool) != SWITCH_STATUS_SUCCESS) {
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing mutex\n");
 			status = SWITCH_STATUS_FALSE;
 			goto done; 
 		}
 
-		cb->interim = interim;
-		strncpy(cb->lang, lang, MAX_LANG);
-
 		/* determine if we need to resample the audio to 16-bit 8khz */
 		if (sampleRate != 8000) {
 			cb->resampler = speex_resampler_init(1, sampleRate, 8000, SWITCH_RESAMPLE_QUALITY, &err);
@@ -461,23 +589,10 @@ extern "C" {
 					switch_channel_get_name(channel), voice_ms, mode);
 			}
 		}
-
-		try {
-			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "%s: initializing gstreamer with %s\n", 
-					switch_channel_get_name(channel), bugname);
-			streamer = new GStreamer(sessionId, bugname, channels, lang, interim, sampleRate, cb->region, subscriptionKey, responseHandler);
-			cb->streamer = streamer;
-			if (!cb->vad) streamer->connect();
-		} catch (std::exception& e) {
-			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing gstreamer: %s.\n", 
-				switch_channel_get_name(channel), e.what());
-			return SWITCH_STATUS_FALSE;
-		}
-
-
-		*ppUserData = cb;
-	
+		if (!cb->vad) streamer->connect();
 	done:
+		*ppUserData = cb;
+		cb->is_keep_alive = 0;
 		return status;
 	}
 
@@ -489,6 +604,12 @@ extern "C" {
 			struct cap_cb *cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
 			switch_status_t st;
 
+			if (use_single_connection && !channelIsClosing) {
+				switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "azure_transcribe_session_stop: call is running, use_single_connection is true, keep alive is activated\n");
+				cb->is_keep_alive = 1;
+				return SWITCH_STATUS_SUCCESS;
+			}
+
 			// close connection and get final responses
 			switch_mutex_lock(cb->mutex);
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "azure_transcribe_session_stop: locked session\n");
@@ -500,6 +621,7 @@ extern "C" {
 			if (streamer) reaper(cb);
 			killcb(cb);
 			switch_mutex_unlock(cb->mutex);
+			switch_mutex_destroy(cb->mutex);
 			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "azure_transcribe_session_stop: unlocked session\n");
 
 			return SWITCH_STATUS_SUCCESS;
@@ -517,7 +639,18 @@ extern "C" {
 
 		frame.data = data;
 		frame.buflen = SWITCH_RECOMMENDED_BUFFER_SIZE;
-
+		if (cb->is_keep_alive) {
+			// remove media bug buffered data
+ 			while (true) {
+				unsigned char data[SWITCH_RECOMMENDED_BUFFER_SIZE] = {0};
+				switch_frame_t frame = { 0 };
+				frame.data = data;
+				frame.buflen = SWITCH_RECOMMENDED_BUFFER_SIZE;
+				switch_status_t rv = switch_core_media_bug_read(bug, &frame, SWITCH_TRUE);
+				if (rv != SWITCH_STATUS_SUCCESS) break;
+			}
+			return SWITCH_TRUE;
+		}
 		if (switch_mutex_trylock(cb->mutex) == SWITCH_STATUS_SUCCESS) {
 			GStreamer* streamer = (GStreamer *) cb->streamer;
 			if (streamer) {

mod_azure_transcribe/mod_azure_transcribe.c

@@ -71,9 +71,9 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	switch_codec_implementation_t read_impl = { 0 };
 	void *pUserData;
 	uint32_t samples_per_second;
-
-
-	if (switch_channel_get_private(channel, bugname)) {
+	int use_single_connection = switch_true(getenv("AZURE_SPEECH_USE_SINGLE_CONNECTION"));
+	bug = switch_channel_get_private(channel, bugname);
+	if (bug && !use_single_connection) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing bug from previous transcribe\n");
 		do_stop(session, bugname);
 	}
@@ -91,12 +91,13 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing azure speech session.\n");
 		return SWITCH_STATUS_FALSE;
 	}
-	if ((status = switch_core_media_bug_add(session, bugname, NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
-		return status;
+	if (!bug || !use_single_connection) {
+		if ((status = switch_core_media_bug_add(session, bugname, NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
+			return status;
+		}
+		switch_channel_set_private(channel, bugname, bug);
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for azure transcribe\n");
 	}
-  switch_channel_set_private(channel, bugname, bug);
-	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for azure transcribe\n");
-
 	return SWITCH_STATUS_SUCCESS;
 }
 

mod_azure_transcribe/mod_azure_transcribe.h

@@ -37,6 +37,8 @@ struct cap_cb {
 	char lang[MAX_LANG];
 	char region[MAX_REGION];
 
+	int is_keep_alive;
+
 	switch_vad_t * vad;
 };
 

mod_azure_tts/Makefile.am

@@ -0,0 +1,10 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_azure_tts
+
+mod_LTLIBRARIES = mod_azure_tts.la
+mod_azure_tts_la_SOURCES  = mod_azure_tts.c azure_glue.cpp
+mod_azure_tts_la_CFLAGS   = $(AM_CFLAGS)
+mod_azure_tts_la_CXXFLAGS = $(AM_CXXFLAGS) -std=c++14 -I/usr/local/include/MicrosoftSpeechSDK/cxx_api -I/usr/local/include/MicrosoftSpeechSDK/c_api
+
+mod_azure_tts_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_azure_tts_la_LDFLAGS  = -avoid-version -module -no-undefined -L/usr/local/lib/MicrosoftSpeechSDK/x64 -lMicrosoft.CognitiveServices.Speech.core -shared -lstdc++ -lboost_system -lboost_thread

mod_azure_tts/azure_glue.cpp

@@ -0,0 +1,414 @@
+#include "mod_azure_tts.h"
+#include <switch.h>
+#include <speechapi_cxx.h>
+
+#include <boost/circular_buffer.hpp>
+
+#include <cstdlib>
+#include <string>
+#include <chrono>
+#include <thread>
+
+#define BUFFER_SIZE 8129
+
+typedef boost::circular_buffer<uint16_t> CircularBuffer_t;
+
+using namespace Microsoft::CognitiveServices::Speech;
+
+static const char* audioLogFile= std::getenv("AZURE_AUDIO_LOGGING");
+
+static std::string fullDirPath;
+
+static void start_synthesis(std::shared_ptr<SpeechSynthesizer> speechSynthesizer, const char* text, azure_t* a) {
+    try {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "start_synthesis calling, text %s\n", text);
+      auto result = std::strncmp(text, "<speak", 6) == 0 ?
+        speechSynthesizer->SpeakSsmlAsync(text).get() :
+        speechSynthesizer->SpeakTextAsync(text).get();
+
+      if (result->Reason == ResultReason::SynthesizingAudioCompleted) {
+        a->response_code = 200;
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "start_synthesis completed id %s, audio data - bytes: %ld, duration: %ldms\n", 
+          result->ResultId.c_str(), result->GetAudioLength(), result->AudioDuration.count());
+      } else if (result->Reason == ResultReason::Canceled) {
+        auto cancellation = SpeechSynthesisCancellationDetails::FromResult(result);
+        a->response_code = static_cast<long int>(cancellation->ErrorCode);
+        a->err_msg = strdup(cancellation->ErrorDetails.c_str());
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error synthesizing text %d with error string: %s.\n",
+          static_cast<int>(cancellation->ErrorCode), cancellation->ErrorDetails.c_str());
+      } else {
+        a->response_code = 500;
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error synthsize text %s (%d).\n", text, static_cast<int>(result->Reason));
+      }
+    } catch (const std::exception& e) {
+        a->response_code = 500;
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "mod_azure_tts: Exception in start_synthesis %s\n",  e.what());
+    }
+    a->draining = 1;
+
+    free((void*) text);
+}
+
+extern "C" {
+  switch_status_t azure_speech_load() {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "azure_speech_loading..\n");
+
+    /* create temp folder for cache files */
+    const char* baseDir = std::getenv("JAMBONZ_TMP_CACHE_FOLDER");
+    if (!baseDir) {
+      baseDir = "/tmp/";
+    }
+    if (strcmp(baseDir, "/") == 0) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", baseDir);
+      return SWITCH_STATUS_FALSE;
+    }
+
+    fullDirPath = std::string(baseDir) + "tts-cache-files";
+
+    // Create the directory with read, write, and execute permissions for everyone
+    mode_t oldMask = umask(0);
+    int result = mkdir(fullDirPath.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
+    umask(oldMask);
+    if (result != 0) {
+      if (errno != EEXIST) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", fullDirPath.c_str());
+        fullDirPath = "";
+      }
+      else switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "folder %s already exists\n", fullDirPath.c_str());
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "created folder %s\n", fullDirPath.c_str());
+    }
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "azure_speech_loaded..\n");
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+    switch_status_t azure_speech_open(azure_t* azure) {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t azure_speech_feed_tts(azure_t* a, char* text, switch_speech_flag_t *flags) {
+    const int MAX_CHARS = 20;
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts %s\n", text);
+
+    /* open cache file */
+    if (a->cache_audio && fullDirPath.length() > 0) {
+      switch_uuid_t uuid;
+      char uuid_str[SWITCH_UUID_FORMATTED_LENGTH + 1];
+      char outfile[512] = "";
+      int fd;
+
+      switch_uuid_get(&uuid);
+      switch_uuid_format(uuid_str, &uuid);
+
+      switch_snprintf(outfile, sizeof(outfile), "%s%s%s.r8", fullDirPath.c_str(), SWITCH_PATH_SEPARATOR, uuid_str);
+      a->cache_filename = strdup(outfile);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "writing audio cache file to %s\n", a->cache_filename);
+
+      mode_t oldMask = umask(0);
+      fd = open(outfile, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
+      umask(oldMask);
+      if (fd == -1 ) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+      }
+      else {
+        a->file = fdopen(fd, "wb");
+        if (!a->file) {
+          close(fd);
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+        }
+      }
+    }
+
+    if (!a->api_key) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "azure_speech_feed_tts: no api_key provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    if (!a->language) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "azure_speech_feed_tts: no language provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    if (a->rate != 8000 /*Hz*/) {
+      int err;
+      a->resampler = speex_resampler_init(1, 8000, a->rate, SWITCH_RESAMPLE_QUALITY, &err);
+      if (0 != err) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing resampler: %s.\n", speex_resampler_strerror(err));
+        return SWITCH_STATUS_FALSE;
+      }
+    }
+
+    std::chrono::time_point<std::chrono::high_resolution_clock>* ptr = new std::chrono::time_point<std::chrono::high_resolution_clock>(std::chrono::high_resolution_clock::now());
+    a->startTime = ptr;
+
+    a->circularBuffer = (void *) new CircularBuffer_t(BUFFER_SIZE);
+
+    auto speechConfig = nullptr != a->endpoint ? 
+			(nullptr != a->api_key ?
+				SpeechConfig::FromEndpoint(a->endpoint, a->api_key) :
+				SpeechConfig::FromEndpoint(a->endpoint)) :
+			SpeechConfig::FromSubscription(a->api_key, a->region ? a->region : "");
+
+    speechConfig->SetSpeechSynthesisOutputFormat(SpeechSynthesisOutputFormat::Raw8Khz16BitMonoPcm);
+    speechConfig->SetSpeechSynthesisLanguage(a->language);
+    speechConfig->SetSpeechSynthesisVoiceName(a->voice_name);
+    if (a->http_proxy_ip) {
+      uint32_t port = a->http_proxy_port && a->http_proxy_port[0] != '\0' ? static_cast<uint32_t>(std::stoul(a->http_proxy_port)) : 80;
+      speechConfig->SetProxy(a->http_proxy_ip, port);
+    }
+
+    if (nullptr != a->endpointId) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts setting endpoint id: %s\n", a->endpointId);
+			speechConfig->SetEndpointId(a->endpointId);
+		}
+
+    if (audioLogFile) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts enabling audio logging to %s\n", audioLogFile);
+      speechConfig->SetProperty(PropertyId::Speech_LogFilename, audioLogFile);
+      speechConfig->EnableAudioLogging();
+    }
+
+    try {
+      auto speechSynthesizer = SpeechSynthesizer::FromConfig(speechConfig, nullptr);
+
+      speechSynthesizer->SynthesisStarted += [a](const SpeechSynthesisEventArgs& e) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts SynthesisStarted\n");
+      };
+
+      speechSynthesizer->Synthesizing += [a](const SpeechSynthesisEventArgs& e) {
+        if (a->flushed) return;
+        bool fireEvent = false;
+        CircularBuffer_t *cBuffer = (CircularBuffer_t *) a->circularBuffer;
+         size_t total_bytes_to_process;
+
+        auto audioData = e.Result->GetAudioData();
+        auto bytes_received = audioData->size();
+        // Buffer to hold combined data if there is unprocessed byte from the last call.
+        std::unique_ptr<uint8_t[]> combinedData;
+        if (a->has_last_byte) {
+          a->has_last_byte = false;  // We'll handle the last_byte now, so toggle the flag off
+
+          // Allocate memory for the new data array
+          combinedData.reset(new uint8_t[bytes_received + 1]);
+
+          // Prepend the last byte from previous call
+          combinedData[0] = a->last_byte;
+
+          // Copy the new data following the prepended byte
+          memcpy(combinedData.get() + 1, audioData->data(), bytes_received);
+
+          total_bytes_to_process = bytes_received + 1;
+        } else {
+          // Allocate memory for the new data array
+          combinedData.reset(new uint8_t[bytes_received]);
+          memcpy(combinedData.get(), audioData->data(), bytes_received);
+          total_bytes_to_process = bytes_received;
+        }
+
+        // If we now have an odd total, save the last byte for next time
+        auto data = combinedData.get();
+        if ((total_bytes_to_process % sizeof(int16_t)) != 0) {
+          a->last_byte = data[total_bytes_to_process - 1];
+          a->has_last_byte = true;
+          total_bytes_to_process--;
+        }
+  
+        if (a->file) {
+          fwrite(data, 1, total_bytes_to_process, a->file);
+        }
+
+        /**
+         * this sort of reinterpretation can be dangerous as a general rule, but in this case we know that the data
+         * is 16-bit PCM, so it's safe to do this and its much faster than copying the data byte by byte
+         */
+        const uint16_t* begin = reinterpret_cast<const uint16_t*>(data);
+        const uint16_t* end = reinterpret_cast<const uint16_t*>(data + total_bytes_to_process);
+
+        /* lock as briefly as possible */
+        switch_mutex_lock(a->mutex);
+        if (cBuffer->capacity() - cBuffer->size() < total_bytes_to_process) {
+          cBuffer->set_capacity(cBuffer->size() + std::max( total_bytes_to_process, (size_t)BUFFER_SIZE));
+        }
+        cBuffer->insert(cBuffer->end(), begin, end);
+        switch_mutex_unlock(a->mutex);
+
+        if (0 == a->reads++) {
+          fireEvent = true;
+        }
+
+        if (fireEvent && a->session_id) {
+          auto endTime = std::chrono::high_resolution_clock::now();
+          auto startTime = *static_cast<std::chrono::time_point<std::chrono::high_resolution_clock>*>(a->startTime);
+          auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(endTime - startTime);
+          auto time_to_first_byte_ms = std::to_string(duration.count());
+          switch_core_session_t* session = switch_core_session_locate(a->session_id);
+          if (session) {
+            switch_channel_t *channel = switch_core_session_get_channel(session);
+            switch_core_session_rwunlock(session);
+            if (channel) {
+              switch_event_t *event;
+              if (switch_event_create(&event, SWITCH_EVENT_PLAYBACK_START) == SWITCH_STATUS_SUCCESS) {
+                switch_channel_event_set_data(channel, event);
+                switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "Playback-File-Type", "tts_stream");
+                switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_time_to_first_byte_ms", time_to_first_byte_ms.c_str());
+                if (a->cache_filename) {
+                  switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", a->cache_filename);
+                }
+                switch_event_fire(&event);
+                a->playback_start_sent = 1;
+              } else {
+                switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "speechSynthesizer->Synthesizing: failed to create event\n");
+              }
+            }else {
+              switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "speechSynthesizer->Synthesizing: channel not found\n");
+            }
+          }
+        }
+      };
+
+      const char* dupText = strdup(text); // text will be freed in the thread
+      std::thread(start_synthesis, speechSynthesizer, dupText, a).detach();
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts sent synthesize request\n");
+    } catch (const std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mod_azure_tts: Exception: %s\n", e.what());
+      return SWITCH_STATUS_FALSE;
+    }
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t azure_speech_read_tts(azure_t* a, void *data, size_t *datalen, switch_speech_flag_t *flags) {
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) a->circularBuffer;
+    std::vector<uint16_t> pcm_data;
+
+    if (a->response_code > 0 && a->response_code != 200) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "azure_speech_read_tts, returning failure\n") ;
+      return SWITCH_STATUS_FALSE;
+    }
+    if (a->flushed) {
+      return SWITCH_STATUS_BREAK;
+    }
+    switch_mutex_lock(a->mutex);
+    size_t bufSize = cBuffer->size();
+    if (cBuffer->empty()) {
+      switch_mutex_unlock(a->mutex);
+      if (a->draining) {
+        return SWITCH_STATUS_BREAK;
+      }
+      /* no audio available yet so send silence */
+      memset(data, 255, *datalen);
+      return SWITCH_STATUS_SUCCESS;
+    }
+    // azure returned 8000hz 16 bit data, we have to take enough data based on call sample rate.
+    size_t size = std::min((*datalen/(2 * a->rate / 8000)), bufSize);
+    pcm_data.insert(pcm_data.end(), cBuffer->begin(), cBuffer->begin() + size);
+    cBuffer->erase(cBuffer->begin(), cBuffer->begin() + size);
+    switch_mutex_unlock(a->mutex);
+
+    size_t data_size = pcm_data.size();
+
+    if (a->resampler) {
+        std::vector<int16_t> in(pcm_data.begin(), pcm_data.end());
+
+        std::vector<int16_t> out((*datalen));
+        spx_uint32_t in_len = data_size;
+        spx_uint32_t out_len = out.size();
+
+        speex_resampler_process_interleaved_int(a->resampler, in.data(), &in_len, out.data(), &out_len);
+
+        if (out_len > out.size()) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, "Resampler output exceeded maximum buffer size!\n");
+          return SWITCH_STATUS_FALSE;
+        }
+
+        memcpy(data, out.data(), out_len * sizeof(int16_t));
+        *datalen = out_len * sizeof(int16_t);
+    } else {
+        memcpy(data, pcm_data.data(), data_size * sizeof(int16_t));
+        *datalen = data_size * sizeof(int16_t);
+    }
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t azure_speech_flush_tts(azure_t* a) {
+    bool download_complete = a->response_code == 200;
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "azure_speech_flush_tts, download complete? %s\n", download_complete ? "yes" : "no") ;
+
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) a->circularBuffer;
+    delete cBuffer;
+    a->circularBuffer = nullptr ;
+    delete static_cast<std::chrono::time_point<std::chrono::high_resolution_clock>*>(a->startTime);
+    a->startTime = nullptr;
+
+    a->flushed = 1;
+    if (!download_complete) {
+      if (a->file) {
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "closing audio cache file %s because download was interrupted\n", a->cache_filename);
+        if (fclose(a->file) != 0) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "error closing audio cache file\n");
+        }
+        a->file = nullptr ;
+      }
+    }
+    // If playback_start has not been sent, delete the file
+    if (a->cache_filename && !a->playback_start_sent) {
+      //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing audio cache file %s because download was interrupted\n", a->cache_filename);
+      if (unlink(a->cache_filename) != 0) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error removing audio cache file %s: %d:%s\n", 
+          a->cache_filename, errno, strerror(errno));
+      }
+      free(a->cache_filename);
+      a->cache_filename = nullptr ;
+    }
+    if (a->session_id) {
+      switch_core_session_t* session = switch_core_session_locate(a->session_id);
+      if (session) {
+        switch_channel_t *channel = switch_core_session_get_channel(session);
+
+        /* unlock as quickly as possible */
+        switch_core_session_rwunlock(session);
+        if (channel) {
+          switch_event_t *event;
+          if (switch_event_create(&event, SWITCH_EVENT_PLAYBACK_STOP) == SWITCH_STATUS_SUCCESS) {
+            switch_channel_event_set_data(channel, event);
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "Playback-File-Type", "tts_stream");
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_azure_response_code", std::to_string(a->response_code).c_str());
+            if (a->cache_filename && download_complete) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", a->cache_filename);
+            }
+            if (!download_complete && a->err_msg) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_error", a->err_msg);
+            }
+            switch_event_fire(&event);
+          }
+          else {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: failed to create event\n");
+          }
+        }
+        else {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: channel not found\n");
+        }
+      }
+    }
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t azure_speech_close(azure_t* a) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "azure_speech_close\n") ;
+    if (a->resampler) {
+      speex_resampler_destroy(a->resampler);
+    }
+
+    a->resampler = NULL;
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t azure_speech_unload() {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+} 

mod_azure_tts/azure_glue.h

@@ -0,0 +1,12 @@
+#ifndef __AZURE_TTS_GLUE_H__
+#define __AZURE_TTS_GLUE_H__
+
+switch_status_t azure_speech_load();
+switch_status_t azure_speech_open(azure_t* azure);
+switch_status_t azure_speech_feed_tts(azure_t* azure, char* text, switch_speech_flag_t *flags);
+switch_status_t azure_speech_read_tts(azure_t* azure, void *data, size_t *datalen, switch_speech_flag_t *flags);
+switch_status_t azure_speech_flush_tts(azure_t* azure);
+switch_status_t azure_speech_close(azure_t* azure);
+switch_status_t azure_speech_unload();
+
+#endif

mod_azure_tts/mod_azure_tts.c

@@ -0,0 +1,177 @@
+#include "mod_azure_tts.h"
+#include "azure_glue.h"
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_azure_tts_load);
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_azure_tts_shutdown);
+SWITCH_MODULE_DEFINITION(mod_azure_tts, mod_azure_tts_load, mod_azure_tts_shutdown, NULL);
+
+static void clearAzure(azure_t* a, int freeAll) {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "clearAzure\n");
+
+  if (a->cache_filename) free(a->cache_filename);
+  if (a->api_key) free(a->api_key);
+  if (a->language) free(a->language);
+  if (a->region) free(a->region);
+  if (a->endpoint) free(a->endpoint);
+  if (a->endpointId) free(a->endpointId);
+  if (a->err_msg) free(a->err_msg);
+  if (a->http_proxy_ip) free(a->http_proxy_ip);
+  if (a->http_proxy_port) free(a->http_proxy_port);
+
+  
+  a->cache_filename = NULL;
+  a->api_key = NULL;
+  a->language = NULL;
+  a->region = NULL;
+  a->endpoint = NULL;
+  a->endpointId = NULL;
+  a->err_msg = NULL;
+  a->http_proxy_ip = NULL;
+  a->http_proxy_port = NULL;
+
+
+  if (freeAll) {
+    if (a->voice_name) free(a->voice_name);
+    if (a->session_id) free(a->session_id);
+    a->voice_name = NULL;
+    a->session_id = NULL;
+  }
+
+}
+
+static azure_t * createOrRetrievePrivateData(switch_speech_handle_t *sh) {
+  azure_t *a = (azure_t *) sh->private_info;  
+  if (!a) {
+    a = switch_core_alloc(sh->memory_pool, sizeof(*a));
+  	sh->private_info = a;
+    memset(a, 0, sizeof(*a));
+    switch_mutex_init(&a->mutex, SWITCH_MUTEX_NESTED, sh->memory_pool);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "allocated azure_t\n");
+  }
+  return a;
+}
+
+switch_status_t a_speech_open(switch_speech_handle_t *sh, const char *voice_name, int rate, int channels, switch_speech_flag_t *flags)
+{
+  azure_t *a = createOrRetrievePrivateData(sh);
+  a->voice_name = strdup(voice_name);
+  a->rate = rate;
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "a_speech_open voice: %s, rate %d, channels %d\n", voice_name, rate, channels);
+  return azure_speech_open(a);
+}
+
+static switch_status_t a_speech_close(switch_speech_handle_t *sh, switch_speech_flag_t *flags)
+{
+  switch_status_t rc;
+  azure_t *a = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "a_speech_close\n");
+
+  switch_mutex_destroy(a->mutex);
+
+  rc = azure_speech_close(a);
+  clearAzure(a, 1);
+  return rc;
+}
+
+/**
+ * Freeswitch will call this function to feed us text to speak
+ */
+static switch_status_t a_speech_feed_tts(switch_speech_handle_t *sh, char *text, switch_speech_flag_t *flags)
+{
+  azure_t *a = createOrRetrievePrivateData(sh);
+  a->draining = 0;
+  a->reads = 0;
+  a->flushed = 0;
+  a->response_code = 0;
+  a->err_msg = NULL;
+  a->has_last_byte = 0;
+  a->playback_start_sent = 0;
+
+  return azure_speech_feed_tts(a, text, flags);
+}
+
+/**
+ * Freeswitch calls periodically to get some rendered audio in L16 format. We can provide up to 8k of audio at a time.
+ */
+static switch_status_t a_speech_read_tts(switch_speech_handle_t *sh, void *data, size_t *datalen, switch_speech_flag_t *flags)
+{
+  azure_t *a = createOrRetrievePrivateData(sh);
+  return azure_speech_read_tts(a, data, datalen, flags);
+}
+
+/**
+ * This is called at the end, not sure exactly what we need to do here..
+ */
+static void a_speech_flush_tts(switch_speech_handle_t *sh)
+{
+  azure_t *a = createOrRetrievePrivateData(sh);
+  azure_speech_flush_tts(a);
+
+  clearAzure(a, 0);
+}
+
+static void a_text_param_tts(switch_speech_handle_t *sh, char *param, const char *val)
+{
+  azure_t *a = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "a_text_param_tts: %s=%s\n", param, val);
+  if (0 == strcmp(param, "api_key")) {
+    if (a->api_key) free(a->api_key);
+    a->api_key = strdup(val);
+  } else if (0 == strcmp(param, "region")) {
+    if (a->region) free(a->region);
+    a->region = strdup(val);
+  } else if (0 == strcmp(param, "voice")) {
+    if (a->voice_name) free(a->voice_name);
+    a->voice_name = strdup(val);
+  } else if (0 == strcmp(param, "language")) {
+    if (a->language) free(a->language);
+    a->language = strdup(val);
+  } else if (0 == strcmp(param, "endpoint")) {
+    if (a->endpoint) free(a->endpoint);
+    a->endpoint = strdup(val);
+  } else if (0 == strcmp(param, "endpointId")) {
+    if (a->endpointId) free(a->endpointId);
+    a->endpointId = strdup(val);
+  } else if (0 == strcmp(param, "http_proxy_ip")) {
+    if (a->http_proxy_ip) free(a->http_proxy_ip);
+    a->http_proxy_ip = strdup(val);
+  } else if (0 == strcmp(param, "http_proxy_port")) {
+    if (a->http_proxy_port) free(a->http_proxy_port);
+    a->http_proxy_port = strdup(val);
+  } else if (0 == strcmp(param, "session-uuid")) {
+    if (a->session_id) free(a->session_id);
+    a->session_id = strdup(val);
+  } else if (0 == strcmp(param, "write_cache_file") && switch_true(val)) {
+    a->cache_audio = 1;
+  }
+}
+
+static void a_numeric_param_tts(switch_speech_handle_t *sh, char *param, int val)
+{
+}
+static void a_float_param_tts(switch_speech_handle_t *sh, char *param, double val)
+{
+}
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_azure_tts_load)
+{
+  switch_speech_interface_t *speech_interface;
+
+  *module_interface = switch_loadable_module_create_module_interface(pool, modname);
+  speech_interface = switch_loadable_module_create_interface(*module_interface, SWITCH_SPEECH_INTERFACE);
+  speech_interface->interface_name = "microsoft";
+  speech_interface->speech_open = a_speech_open;
+  speech_interface->speech_close = a_speech_close;
+  speech_interface->speech_feed_tts = a_speech_feed_tts;
+  speech_interface->speech_read_tts = a_speech_read_tts;
+	speech_interface->speech_flush_tts = a_speech_flush_tts;
+	speech_interface->speech_text_param_tts = a_text_param_tts;
+	speech_interface->speech_numeric_param_tts = a_numeric_param_tts;
+	speech_interface->speech_float_param_tts = a_float_param_tts;
+  return azure_speech_load();
+}
+
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_azure_tts_shutdown)
+{
+  return azure_speech_unload();
+}

mod_azure_tts/mod_azure_tts.h

@@ -0,0 +1,41 @@
+#ifndef __MOD_AZURE_TTS_H__
+#define __MOD_AZURE_TTS_H__
+
+#include <switch.h>
+#include <speex/speex_resampler.h>
+
+typedef struct azure_data {
+  char *voice_name;
+  char *api_key;
+  char *region;
+  char *language;
+  char *endpoint;
+  char *endpointId;
+  char *http_proxy_ip;
+  char *http_proxy_port;
+
+  /* result data */
+  long response_code;
+  char *session_id;
+  char *cache_filename;
+  char *err_msg;
+
+  int rate;
+  int draining;
+  int reads;
+  int cache_audio;
+  int flushed;
+  int playback_start_sent;
+
+  void *startTime;
+
+  FILE *file;
+  SpeexResamplerState *resampler;
+  void *circularBuffer;
+  switch_mutex_t *mutex;
+
+  int has_last_byte;
+  uint8_t last_byte;
+} azure_t;
+
+#endif

mod_cobalt_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_custom_tts/LICENSE

@@ -0,0 +1,8 @@
+Copyright 2023, Drachtio Communications Services, LLC
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+

mod_custom_tts/Makefile.am

@@ -0,0 +1,8 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_custom_tts
+
+mod_LTLIBRARIES = mod_custom_tts.la
+mod_custom_tts_la_SOURCES  = mod_custom_tts.c custom_glue.cpp
+mod_custom_tts_la_CFLAGS   = $(AM_CFLAGS)
+mod_custom_tts_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_custom_tts_la_LDFLAGS  = -avoid-version -module -no-undefined -shared -lstdc++ -lboost_system -lboost_thread -lmpg123

mod_custom_tts/README.md

@@ -0,0 +1,3 @@
+# mod_custom_tts
+
+A Freeswitch module that allows speak text to speech audio from custom vendor stream.

mod_custom_tts/custom_glue.cpp

@@ -0,0 +1,981 @@
+#include "mod_custom_tts.h"
+#include <switch.h>
+#include <switch_json.h>
+#include <curl/curl.h>
+#include <cstdlib>
+
+#include <boost/circular_buffer.hpp>
+#include <boost/thread.hpp>
+#include <boost/asio.hpp>
+#include <boost/asio/ssl.hpp>
+#include <boost/pool/object_pool.hpp>
+#include <boost/bind/bind.hpp>
+#include <boost/tokenizer.hpp>
+#include <boost/foreach.hpp>
+#include <boost/asio.hpp>
+#include <boost/assign/list_of.hpp>
+#include <boost/algorithm/string.hpp>
+
+#include "mpg123.h"
+
+#define BUFFER_GROW_SIZE (8192)
+#define MP3_DCACHE 8192 * 2
+
+typedef boost::circular_buffer<uint16_t> CircularBuffer_t;
+/* Global information, common to all connections */
+typedef struct
+{
+  CURLM *multi;
+  int still_running;
+} GlobalInfo_t;
+static GlobalInfo_t global;
+
+/* Information associated with a specific easy handle */
+typedef struct
+{
+  CURL *easy;
+  custom_t* custom;
+  char* body;
+  struct curl_slist *hdr_list;
+  GlobalInfo_t *global;
+  mpg123_handle *mh;
+  char error[CURL_ERROR_SIZE];
+  FILE* file;
+  std::chrono::time_point<std::chrono::high_resolution_clock> startTime;
+  bool flushed;
+} ConnInfo_t;
+
+
+static boost::object_pool<ConnInfo_t> pool ;
+static std::map<curl_socket_t, boost::asio::ip::tcp::socket *> socket_map;
+static boost::asio::io_service io_service;
+static boost::asio::deadline_timer timer(io_service);
+static std::string fullDirPath;
+static std::thread worker_thread;
+
+std::string secondsToMillisecondsString(double seconds) {
+    // Convert to milliseconds
+    double milliseconds = seconds * 1000.0;
+
+    // Truncate to remove fractional part
+    long milliseconds_long = static_cast<long>(milliseconds);
+
+    // Convert to string
+    return std::to_string(milliseconds_long);
+}
+
+static std::string getEnvVar(const std::string& varName) {
+    const char* val = std::getenv(varName.c_str());
+    return val ? std::string(val) : "";
+}
+
+static long getConnectionTimeout() {
+    std::string connectTimeoutStr = getEnvVar("CUSTOM_TTS_CURL_CONNECT_TIMEOUT");
+
+    if (connectTimeoutStr.empty()) {
+        connectTimeoutStr = getEnvVar("TTS_CURL_CONNECT_TIMEOUT");
+    }
+
+    long connectTimeout = 10L;
+
+    if (!connectTimeoutStr.empty()) {
+        connectTimeout = std::stol(connectTimeoutStr);
+    }
+
+    return connectTimeout;
+}
+
+static CURL* createEasyHandle(void) {
+  CURL* easy = curl_easy_init();
+  if(!easy) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "curl_easy_init() failed!\n");
+    return nullptr ;
+  }  
+
+  curl_easy_setopt(easy, CURLOPT_FOLLOWLOCATION, 1L);
+  curl_easy_setopt(easy, CURLOPT_USERAGENT, "jambonz/0.8.5");
+
+  // set connect timeout to 3 seconds and total timeout to 109 seconds
+  curl_easy_setopt(easy, CURLOPT_CONNECTTIMEOUT_MS, 3000L);
+  curl_easy_setopt(easy, CURLOPT_TIMEOUT, getConnectionTimeout());
+
+  return easy ;    
+}
+
+static void cleanupConn(ConnInfo_t *conn) {
+  auto c = conn->custom;
+
+  if (conn->mh) {
+    mpg123_close(conn->mh);
+    mpg123_delete(conn->mh);
+  }
+
+  if( conn->hdr_list ) {
+    curl_slist_free_all(conn->hdr_list);
+    conn->hdr_list = nullptr ;
+  }
+  curl_easy_cleanup(conn->easy);
+
+  if (conn->file) {
+    if (fclose(conn->file) != 0) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error closing audio cache file\n");
+    }
+    conn->file = nullptr ;
+  }
+
+  c->conn = nullptr ;
+  c->draining = 1;
+
+  memset(conn, 0, sizeof(ConnInfo_t));
+  pool.destroy(conn) ;
+}
+
+/* Check for completed transfers, and remove their easy handles */
+void check_multi_info(GlobalInfo_t *g) {
+  CURLMsg *msg;
+  int msgs_left;
+  ConnInfo_t *conn;
+  CURL *easy;
+  CURLcode res;
+  
+  while((msg = curl_multi_info_read(g->multi, &msgs_left))) {
+    if(msg->msg == CURLMSG_DONE) {
+      long response_code;
+      double namelookup=0, connect=0, total=0 ;
+      char *ct = NULL ;
+
+      easy = msg->easy_handle;
+      res = msg->data.result;
+      curl_easy_getinfo(easy, CURLINFO_PRIVATE, &conn);
+      curl_easy_getinfo(easy, CURLINFO_RESPONSE_CODE, &response_code);
+      curl_easy_getinfo(easy, CURLINFO_CONTENT_TYPE, &ct);
+
+      curl_easy_getinfo(easy, CURLINFO_NAMELOOKUP_TIME, &namelookup);
+      curl_easy_getinfo(easy, CURLINFO_CONNECT_TIME, &connect);
+      curl_easy_getinfo(easy, CURLINFO_TOTAL_TIME, &total);
+
+      auto c = conn->custom;
+      c->response_code = response_code;
+      if (ct) c->ct = strdup(ct);
+
+      std::string name_lookup_ms = secondsToMillisecondsString(namelookup);
+      std::string connect_ms = secondsToMillisecondsString(connect);
+      std::string final_response_time_ms = secondsToMillisecondsString(total);
+
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, 
+        "mod_custom_tts: response: %ld, content-type %s,"
+        "dns(ms): %"  CURL_FORMAT_CURL_OFF_T ".%06ld, "
+        "connect(ms): %"  CURL_FORMAT_CURL_OFF_T ".%06ld, "
+        "total(ms): %"  CURL_FORMAT_CURL_OFF_T ".%06ld\n",
+        response_code, ct,
+        (long)(namelookup), (long)(fmod(namelookup, 1.0) * 1000000),
+        (long)(connect), (long)(fmod(connect, 1.0) * 1000000),
+        (long)(total), (long)(fmod(total, 1.0) * 1000000));
+
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "name lookup time: %s\n", name_lookup_ms.c_str());
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "connect time: %s\n", connect_ms.c_str());
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "final response time: %s\n", final_response_time_ms.c_str());
+
+      c->name_lookup_time_ms = strdup(name_lookup_ms.c_str());
+      c->connect_time_ms = strdup(connect_ms.c_str());
+      c->final_response_time_ms = strdup(final_response_time_ms.c_str());
+
+      curl_multi_remove_handle(g->multi, easy);
+      cleanupConn(conn);
+    }
+  }
+}
+
+int mcode_test(const char *where, CURLMcode code) {
+  if(CURLM_OK != code) {
+    const char *s;
+    switch(code) {
+    case CURLM_CALL_MULTI_PERFORM:
+      s = "CURLM_CALL_MULTI_PERFORM";
+      break;
+    case CURLM_BAD_HANDLE:
+      s = "CURLM_BAD_HANDLE";
+      break;
+    case CURLM_BAD_EASY_HANDLE:
+      s = "CURLM_BAD_EASY_HANDLE";
+      break;
+    case CURLM_OUT_OF_MEMORY:
+      s = "CURLM_OUT_OF_MEMORY";
+      break;
+    case CURLM_INTERNAL_ERROR:
+      s = "CURLM_INTERNAL_ERROR";
+      break;
+    case CURLM_UNKNOWN_OPTION:
+      s = "CURLM_UNKNOWN_OPTION";
+      break;
+    case CURLM_LAST:
+      s = "CURLM_LAST";
+      break;
+    default:
+      s = "CURLM_unknown";
+      break;
+    case CURLM_BAD_SOCKET:
+      s = "CURLM_BAD_SOCKET";
+      break;
+    }
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mcode_test ERROR: %s returns %s:%d\n", where, s, code);
+
+    return -1;
+  }
+  return 0 ;
+}
+
+static void remsock(int *f, GlobalInfo_t *g) {
+  if(f) {
+    free(f);
+    f = NULL;
+  }
+}
+
+/* Called by asio when there is an action on a socket */
+static void event_cb(GlobalInfo_t *g, curl_socket_t s, int action, const boost::system::error_code & error, int *fdp) {
+  int f = *fdp;
+
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "event_cb socket %#X has action %d\n", s, action) ; 
+
+  // Socket already POOL REMOVED.
+  if (f == CURL_POLL_REMOVE) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "event_cb socket %#X removed\n", s); 
+    remsock(fdp, g);
+    return;
+  }
+
+  if(socket_map.find(s) == socket_map.end()) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "event_cb: socket  %#X already closed\n, s");
+    return;
+  }
+
+  /* make sure the event matches what are wanted */
+  if(f == action || f == CURL_POLL_INOUT) {
+    if(error) {
+      action = CURL_CSELECT_ERR;
+    }
+    CURLMcode rc = curl_multi_socket_action(g->multi, s, action, &g->still_running);
+
+    mcode_test("event_cb: curl_multi_socket_action", rc);
+    check_multi_info(g);
+
+    if(g->still_running <= 0) {
+      timer.cancel();
+    }
+
+    /* keep on watching.
+      * the socket may have been closed and/or fdp may have been changed
+      * in curl_multi_socket_action(), so check them both */
+    if(!error && socket_map.find(s) != socket_map.end() &&
+        (f == action || f == CURL_POLL_INOUT)) {
+      boost::asio::ip::tcp::socket *tcp_socket = socket_map.find(s)->second;
+
+      if(action == CURL_POLL_IN) {
+        tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                                action, boost::placeholders::_1, fdp));
+      }
+      if(action == CURL_POLL_OUT) {
+        tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                      boost::bind(&event_cb, g, s,
+                                                  action, boost::placeholders::_1, fdp));
+      } 
+    }
+  }
+}
+
+/* socket functions */
+static void setsock(int *fdp, curl_socket_t s, CURL *e, int act, int oldact, GlobalInfo_t *g) {
+  std::map<curl_socket_t, boost::asio::ip::tcp::socket *>::iterator it = socket_map.find(s);
+
+  if(it == socket_map.end()) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "setsock: socket  %#X not found\n, s");
+    return;
+  }
+
+  boost::asio::ip::tcp::socket * tcp_socket = it->second;
+
+  *fdp = act;
+
+  if(act == CURL_POLL_IN) {
+    if(oldact != CURL_POLL_IN && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                  boost::bind(&event_cb, g, s,
+                                  CURL_POLL_IN, boost::placeholders::_1, fdp));
+    }
+  }
+  else if(act == CURL_POLL_OUT) {
+    if(oldact != CURL_POLL_OUT && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                    CURL_POLL_OUT, boost::placeholders::_1, fdp));
+    }
+  }
+  else if(act == CURL_POLL_INOUT) {
+    if(oldact != CURL_POLL_IN && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                  boost::bind(&event_cb, g, s,
+                                  CURL_POLL_IN, boost::placeholders::_1, fdp));
+    }
+    if(oldact != CURL_POLL_OUT && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                    CURL_POLL_OUT, boost::placeholders::_1, fdp));
+    }
+  }
+}
+
+static void addsock(curl_socket_t s, CURL *easy, int action, GlobalInfo_t *g) {
+  /* fdp is used to store current action */
+  int *fdp = (int *) calloc(sizeof(int), 1);
+
+  setsock(fdp, s, easy, action, 0, g);
+  curl_multi_assign(g->multi, s, fdp);
+}
+
+static int sock_cb(CURL *e, curl_socket_t s, int what, void *cbp, void *sockp) {
+  GlobalInfo_t *g = &global;
+
+  int *actionp = (int *) sockp;
+  static const char *whatstr[] = { "none", "IN", "OUT", "INOUT", "REMOVE"};
+
+  if(what == CURL_POLL_REMOVE) {
+    *actionp = what;
+  }
+  else {
+    if(!actionp) {
+      addsock(s, e, what, g);
+    }
+    else {
+      setsock(actionp, s, e, what, *actionp, g);
+    }
+  }
+  return 0;  
+}
+
+static void threadFunc() {      
+  /* to make sure the event loop doesn't terminate when there is no work to do */
+  io_service.reset() ;
+  boost::asio::io_service::work work(io_service);
+  
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_custom_tts threadFunc - starting\n");
+
+  for(;;) {
+      
+    try {
+      io_service.run() ;
+      break ;
+    }
+    catch( std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mod_custom_tts threadFunc - Error: %s\n", e.what());
+    }
+  }
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_custom_tts threadFunc - ending\n");
+}
+
+
+/* Called by asio when our timeout expires */
+static void timer_cb(const boost::system::error_code & error, GlobalInfo_t *g)
+{
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "timer_cb\n");
+
+  if(!error) {
+    CURLMcode rc = curl_multi_socket_action(g->multi, CURL_SOCKET_TIMEOUT, 0, &g->still_running);
+    mcode_test("timer_cb: curl_multi_socket_action", rc);
+    check_multi_info(g);
+  }
+}
+
+int multi_timer_cb(CURLM *multi, long timeout_ms, GlobalInfo_t *g) {
+
+  /* cancel running timer */
+  timer.cancel();
+
+  if(timeout_ms >= 0) {
+    // from libcurl 7.88.1-10+deb12u4 does not allow call curl_multi_socket_action or curl_multi_perform in curl_multi callback directly
+    timer.expires_from_now(boost::posix_time::millisec(timeout_ms ? timeout_ms : 1));
+    timer.async_wait(boost::bind(&timer_cb, boost::placeholders::_1, g));
+  }
+
+  return 0;
+}
+
+static std::vector<uint16_t> convert_mp3_to_linear(ConnInfo_t *conn, uint8_t *data, size_t len) {
+  std::vector<uint16_t> linear_data;
+  int eof = 0;
+  int mp3err = 0;
+  unsigned char decode_buf[MP3_DCACHE];
+
+  if(mpg123_feed(conn->mh, data, len) == MPG123_OK) {
+    while(!eof) {
+      size_t usedlen = 0;
+      off_t frame_offset;
+      unsigned char* audio;
+
+      int decode_status = mpg123_decode_frame(conn->mh, &frame_offset, &audio, &usedlen);
+
+      switch(decode_status) {
+        case MPG123_NEW_FORMAT:
+          continue;
+
+        case MPG123_OK:
+          for(size_t i = 0; i < usedlen; i += 2) {
+            uint16_t value = reinterpret_cast<uint16_t*>(audio)[i / 2];
+            linear_data.push_back(value);
+          }
+          break;
+
+        case MPG123_DONE:
+        case MPG123_NEED_MORE:
+          eof = 1;
+          break;
+
+        case MPG123_ERR:
+        default:
+          if(++mp3err >= 5) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Decoder Error!\n");
+            eof = 1;
+          }
+      }
+
+      if (eof)
+        break;
+
+      mp3err = 0;
+    }
+  }
+
+  return linear_data;
+}
+/* CURLOPT_WRITEFUNCTION */
+static size_t write_cb(void *ptr, size_t size, size_t nmemb, ConnInfo_t *conn) {
+  bool fireEvent = false;
+  uint8_t *data = (uint8_t *) ptr;
+  size_t bytes_received = size * nmemb;
+  auto c = conn->custom;
+  CircularBuffer_t *cBuffer = (CircularBuffer_t *) c->circularBuffer;
+  std::vector<uint16_t> pcm_data;
+  
+  if (conn->flushed || cBuffer == nullptr) {
+    /* this will abort the transfer */
+    return 0;
+  }
+  {
+    switch_mutex_lock(c->mutex);
+
+    if (c->response_code > 0 && c->response_code != 200) {
+      std::string body((char *) ptr, bytes_received);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: received body %s\n", body.c_str());
+      c->err_msg = strdup(body.c_str());
+      switch_mutex_unlock(c->mutex);
+      return 0;
+    }
+
+    pcm_data = convert_mp3_to_linear(conn, data, bytes_received);
+    size_t bytesResampled = pcm_data.size() * sizeof(uint16_t);
+
+    /* cache same data to avoid streaming and cached audio quality is different*/
+    if (conn->file) fwrite(pcm_data.data(), sizeof(uint8_t), bytesResampled, conn->file);
+
+    // Resize the buffer if necessary
+    if (cBuffer->capacity() - cBuffer->size() < (bytesResampled / sizeof(uint16_t))) {
+      //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "write_cb growing buffer\n"); 
+
+      //TODO: if buffer exceeds some max size, return CURL_WRITEFUNC_ERROR to abort the transfer
+      cBuffer->set_capacity(cBuffer->size() + std::max((bytesResampled / sizeof(uint16_t)), (size_t)BUFFER_GROW_SIZE));
+    }
+    
+    /* Push the data into the buffer */
+    cBuffer->insert(cBuffer->end(), pcm_data.data(), pcm_data.data() + pcm_data.size());
+
+    if (0 == c->reads++) {
+      fireEvent = true;
+    }
+    switch_mutex_unlock(c->mutex);
+  }
+  if (fireEvent && c->session_id) {
+    auto endTime = std::chrono::high_resolution_clock::now();
+    auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(endTime - conn->startTime);
+    auto time_to_first_byte_ms = std::to_string(duration.count());
+    switch_core_session_t* session = switch_core_session_locate(c->session_id);
+    if (session) {
+      switch_channel_t *channel = switch_core_session_get_channel(session);
+      switch_core_session_rwunlock(session);
+      if (channel) {
+        switch_event_t *event;
+        if (switch_event_create(&event, SWITCH_EVENT_PLAYBACK_START) == SWITCH_STATUS_SUCCESS) {
+          switch_channel_event_set_data(channel, event);
+
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "write_cb: firing playback-started\n");
+
+          switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "Playback-File-Type", "tts_stream");
+          if (c->name_lookup_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_custom_name_lookup_time_ms", c->name_lookup_time_ms);
+          }
+          if (c->connect_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_custom_connect_time_ms", c->connect_time_ms);
+          }
+          if (c->final_response_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_custom_final_response_time_ms", c->final_response_time_ms);
+          }
+          if (c->voice_name) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_custom_voice_name", c->voice_name);
+          }
+          if (c->cache_filename) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", c->cache_filename);
+          }
+
+          switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_time_to_first_byte_ms", time_to_first_byte_ms.c_str());
+          switch_event_fire(&event);
+        }
+        else {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: failed to create event\n");
+        }
+      }
+      else {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: channel not found\n");
+      }
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: session %s not found\n", c->session_id);
+    }
+  }
+  return bytes_received;
+}
+
+static bool parseHeader(const std::string& str, std::string& header, std::string& value) {
+    std::vector<std::string> parts;
+    boost::split(parts, str, boost::is_any_of(":"), boost::token_compress_on);
+
+    if (parts.size() != 2)
+        return false;
+
+    header = boost::trim_copy(parts[0]);
+    value = boost::trim_copy(parts[1]);
+    return true;
+}
+
+static int extract_response_code(const std::string& input) {
+  std::size_t space_pos = input.find(' ');
+  if (space_pos == std::string::npos) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Invalid HTTP response format %s\n", input.c_str());
+    return 0;
+  }
+
+  std::size_t code_start_pos = space_pos + 1;
+  std::size_t code_end_pos = input.find(' ', code_start_pos);
+  if (code_end_pos == std::string::npos) {
+    code_end_pos = input.length();
+  }
+
+  std::string code_str = input.substr(code_start_pos, code_end_pos - code_start_pos);
+  int response_code = std::stoi(code_str);
+  return response_code;
+}
+
+static size_t header_callback(char *buffer, size_t size, size_t nitems, ConnInfo_t *conn) {
+  size_t bytes_received = size * nitems;
+  const std::string prefix = "HTTP/";
+  custom_t* c = conn->custom;
+  std::string header, value;
+  std::string input(buffer, bytes_received);
+  if (parseHeader(input, header, value)) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "recv header: %s with value %s\n", header.c_str(), value.c_str());
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: %s\n", input.c_str());
+    if (input.rfind(prefix, 0) == 0) {
+      try {
+        c->response_code = extract_response_code(input);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: parsed response code: %ld\n", c->response_code);
+      } catch (const std::invalid_argument& e) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "header_callback: invalid response code %s\n", input.substr(prefix.length()).c_str());
+      }
+    }
+  }
+  return bytes_received;
+}
+
+/* CURLOPT_OPENSOCKETFUNCTION */
+static curl_socket_t opensocket(void *clientp, curlsocktype purpose, struct curl_sockaddr *address) {
+  curl_socket_t sockfd = CURL_SOCKET_BAD;
+
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "opensocket: %d\n", purpose);
+  /* restrict to IPv4 */
+  if(purpose == CURLSOCKTYPE_IPCXN && address->family == AF_INET) {
+    /* create a tcp socket object */
+    boost::asio::ip::tcp::socket *tcp_socket = new boost::asio::ip::tcp::socket(io_service);
+
+    /* open it and get the native handle*/
+    boost::system::error_code ec;
+    tcp_socket->open(boost::asio::ip::tcp::v4(), ec);
+
+    if(ec) {
+      /* An error occurred */
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't open socket [%ld][%s]\n", ec, ec.message().c_str());
+    }
+    else {
+      sockfd = tcp_socket->native_handle();
+
+      /* save it for monitoring */
+      socket_map.insert(std::pair<curl_socket_t, boost::asio::ip::tcp::socket *>(sockfd, tcp_socket));
+    }
+  }
+  return sockfd;
+}
+
+/* CURLOPT_CLOSESOCKETFUNCTION */
+static int close_socket(void *clientp, curl_socket_t item) {
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "close_socket : %#X\n", item);
+
+  std::map<curl_socket_t, boost::asio::ip::tcp::socket *>::iterator it = socket_map.find(item);
+  if(it != socket_map.end()) {
+    delete it->second;
+    socket_map.erase(it);
+  }
+  return 0;
+}
+
+
+extern "C" {
+  switch_status_t custom_speech_load() {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "custom_speech_loading..\n");
+    memset(&global, 0, sizeof(GlobalInfo_t));
+    global.multi = curl_multi_init();
+
+     if (!global.multi) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "custom_speech_load curl_multi_init() failed, exiting!\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    curl_multi_setopt(global.multi, CURLMOPT_SOCKETFUNCTION, sock_cb);
+    curl_multi_setopt(global.multi, CURLMOPT_SOCKETDATA, &global);
+    curl_multi_setopt(global.multi, CURLMOPT_TIMERFUNCTION, multi_timer_cb);
+    curl_multi_setopt(global.multi, CURLMOPT_TIMERDATA, &global);
+    curl_multi_setopt(global.multi, CURLMOPT_PIPELINING, CURLPIPE_MULTIPLEX);
+
+    /* create temp folder for cache files */
+    const char* baseDir = std::getenv("JAMBONZ_TMP_CACHE_FOLDER");
+    if (!baseDir) {
+      baseDir = "/tmp/";
+    }
+    if (strcmp(baseDir, "/") == 0) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", baseDir);
+      return SWITCH_STATUS_FALSE;
+    }
+
+    fullDirPath = std::string(baseDir) + "tts-cache-files";
+
+    // Create the directory with read, write, and execute permissions for everyone
+    mode_t oldMask = umask(0);
+    int result = mkdir(fullDirPath.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
+    umask(oldMask);
+    if (result != 0) {
+      if (errno != EEXIST) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", fullDirPath.c_str());
+        fullDirPath = "";
+      }
+      else switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "folder %s already exists\n", fullDirPath.c_str());
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "created folder %s\n", fullDirPath.c_str());
+    }
+    // init mgp123
+    if (mpg123_init() != MPG123_OK) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to initiate MPG123");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    /* start worker thread that handles transfers*/
+    std::thread t(threadFunc) ;
+    worker_thread.swap( t ) ;
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "custom_speech_loaded..\n");
+
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t custom_speech_unload() {
+    /* stop the ASIO IO service */
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "custom_speech_unload: stopping io service\n");
+    io_service.stop();
+
+    /* Join the worker thread */
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "custom_speech_unload: wait for worker thread to complete\n");
+    if (worker_thread.joinable()) {
+        worker_thread.join();
+    }
+
+    /* cleanup curl multi handle*/
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "custom_speech_unload: release curl multi\n");
+    curl_multi_cleanup(global.multi);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "custom_speech_unload: completed\n");
+
+    mpg123_exit();
+
+		return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t custom_speech_open(custom_t* custom) {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t custom_speech_feed_tts(custom_t* c, char* text, switch_speech_flag_t *flags) {
+    CURLMcode rc;
+
+    const int MAX_CHARS = 20;
+    char tempText[MAX_CHARS + 4]; // +4 for the ellipsis and null terminator
+
+    if (strlen(text) > MAX_CHARS) {
+        strncpy(tempText, text, MAX_CHARS);
+        strcpy(tempText + MAX_CHARS, "...");
+    } else {
+        strcpy(tempText, text);
+    }
+
+    /* open cache file */
+    if (c->cache_audio && fullDirPath.length() > 0) {
+      switch_uuid_t uuid;
+      char uuid_str[SWITCH_UUID_FORMATTED_LENGTH + 1];
+      char outfile[512] = "";
+      int fd;
+
+      switch_uuid_get(&uuid);
+      switch_uuid_format(uuid_str, &uuid);
+
+      switch_snprintf(outfile, sizeof(outfile), "%s%s%s.mp3", fullDirPath.c_str(), SWITCH_PATH_SEPARATOR, uuid_str);
+      c->cache_filename = strdup(outfile);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "writing audio cache file to %s\n", c->cache_filename);
+
+      mode_t oldMask = umask(0);
+      fd = open(outfile, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
+      umask(oldMask);
+      if (fd == -1 ) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+      }
+      else {
+        c->file = fdopen(fd, "wb");
+        if (!c->file) {
+          close(fd);
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+        }
+      }
+    }
+
+    if (!c->custom_tts_url) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "custom_speech_feed_tts: no custom_tts_url provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    /* format url*/
+    std::string url = c->custom_tts_url;
+
+    /* create the JSON body */
+    cJSON * jResult = cJSON_CreateObject();
+    cJSON_AddStringToObject(jResult, "text", text);
+    cJSON_AddStringToObject(jResult, "type", std::strncmp(text, "<speak", 6) == 0 ? "ssml" : "text");
+    if (c->language) {
+      cJSON_AddStringToObject(jResult, "language", c->language);
+    }
+    if (c->voice_name) {
+      cJSON_AddStringToObject(jResult, "voice", c->voice_name);
+    }
+    char *json = cJSON_PrintUnformatted(jResult);
+
+    cJSON_Delete(jResult);
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "custom_speech_feed_tts: [%s] [%s]\n", url.c_str(), tempText);
+
+    ConnInfo_t *conn = pool.malloc() ;
+
+    // COnfigure MPG123
+    int mhError = 0;
+    mpg123_handle *mh = mpg123_new("auto", &mhError);
+    if (!mh) {
+      const char *mhErr = mpg123_plain_strerror(mhError);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error allocating mpg123 handle! %s\n", switch_str_nil(mhErr));
+      return SWITCH_STATUS_FALSE;
+    }
+
+    if (mpg123_open_feed(mh) != MPG123_OK) {
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error mpg123_open_feed!\n");
+      return SWITCH_STATUS_FALSE;
+		}
+
+    if (mpg123_format_all(mh) != MPG123_OK) {
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error mpg123_format_all!\n");
+      return SWITCH_STATUS_FALSE;
+		}
+
+    if (mpg123_param(mh, MPG123_FLAGS, MPG123_MONO_MIX, 0) != MPG123_OK) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error forcing single channel!\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    CURL* easy = createEasyHandle();
+    c->conn = (void *) conn ;
+    conn->custom = c;
+    conn->easy = easy;
+    conn->mh = mh;
+    conn->global = &global;
+    conn->hdr_list = NULL ;
+    conn->file = c->file;
+    conn->body = json;
+    conn->flushed = false;
+    
+
+    c->circularBuffer = (void *) new CircularBuffer_t(8192);
+
+    if (mpg123_param(mh, MPG123_FORCE_RATE, c->rate /*Hz*/, 0) != MPG123_OK) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error mpg123_param!\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    curl_easy_setopt(easy, CURLOPT_URL, url.c_str());
+    curl_easy_setopt(easy, CURLOPT_WRITEFUNCTION, write_cb);
+    curl_easy_setopt(easy, CURLOPT_WRITEDATA, conn);
+    curl_easy_setopt(easy, CURLOPT_ERRORBUFFER, conn->error);
+    curl_easy_setopt(easy, CURLOPT_PRIVATE, conn);
+    curl_easy_setopt(easy, CURLOPT_VERBOSE, 0L);
+    curl_easy_setopt(easy, CURLOPT_NOPROGRESS, 1L);
+    curl_easy_setopt(easy, CURLOPT_HEADERFUNCTION, header_callback);
+    curl_easy_setopt(easy, CURLOPT_HEADERDATA, conn);
+    
+    /* call this function to get a socket */
+    curl_easy_setopt(easy, CURLOPT_OPENSOCKETFUNCTION, opensocket);
+
+    /* call this function to close a socket */
+    curl_easy_setopt(easy, CURLOPT_CLOSESOCKETFUNCTION, close_socket);
+
+    if (c->auth_token) {
+      std::ostringstream api_key_stream;
+      api_key_stream << "Authorization: Bearer " << c->auth_token;
+      conn->hdr_list = curl_slist_append(conn->hdr_list, api_key_stream.str().c_str());
+    }
+    conn->hdr_list = curl_slist_append(conn->hdr_list, "Accept: audio/mpeg");
+    conn->hdr_list = curl_slist_append(conn->hdr_list, "Content-Type: application/json");
+    curl_easy_setopt(easy, CURLOPT_HTTPHEADER, conn->hdr_list);
+
+    curl_easy_setopt(easy, CURLOPT_POSTFIELDS, conn->body);
+    //curl_easy_setopt(easy, CURLOPT_POSTFIELDSIZE, body.length());
+
+    curl_easy_setopt(easy, CURLOPT_HTTP_VERSION, CURL_HTTP_VERSION_2_0);
+
+    rc = curl_multi_add_handle(global.multi, conn->easy);
+    mcode_test("new_conn: curl_multi_add_handle", rc);
+
+    /* start a timer to measure the duration until we receive first byte of audio */
+    conn->startTime = std::chrono::high_resolution_clock::now();
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "custom_speech_feed_tts: called curl_multi_add_handle\n");
+
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t custom_speech_read_tts(custom_t* c, void *data, size_t *datalen, switch_speech_flag_t *flags) {
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) c->circularBuffer;
+    std::vector<uint16_t> pcm_data;
+
+    {
+      switch_mutex_lock(c->mutex);
+      ConnInfo_t *conn = (ConnInfo_t *) c->conn;
+      if (c->response_code > 0 && c->response_code != 200) {
+        switch_mutex_unlock(c->mutex);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "custom_speech_read_tts, returning failure\n") ;  
+        return SWITCH_STATUS_FALSE;
+      }
+      if (conn && conn->flushed) {
+        switch_mutex_unlock(c->mutex);
+        return SWITCH_STATUS_BREAK;
+      }
+      if (cBuffer->empty()) {
+        if (c->draining) {
+          switch_mutex_unlock(c->mutex);
+          return SWITCH_STATUS_BREAK;
+        }
+        /* no audio available yet so send silence */
+        memset(data, 255, *datalen);
+        switch_mutex_unlock(c->mutex);
+        return SWITCH_STATUS_SUCCESS;
+      }
+      size_t size = std::min((*datalen/2), cBuffer->size());
+      pcm_data.insert(pcm_data.end(), cBuffer->begin(), cBuffer->begin() + size);
+      cBuffer->erase(cBuffer->begin(), cBuffer->begin() + size);
+      switch_mutex_unlock(c->mutex);
+    }
+
+    memcpy(data, pcm_data.data(), pcm_data.size() * sizeof(uint16_t));
+    *datalen = pcm_data.size() * sizeof(uint16_t);
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t custom_speech_flush_tts(custom_t* c) {
+    bool download_complete = c->response_code == 200;
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "custom_speech_flush_tts, download complete? %s\n", download_complete ? "yes" : "no") ;  
+    ConnInfo_t *conn = (ConnInfo_t *) c->conn;
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) c->circularBuffer;
+    delete cBuffer;
+    c->circularBuffer = nullptr ;
+
+    if (conn) {
+      conn->flushed = true;
+      if (!download_complete) {
+        if (conn->file) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "closing audio cache file %s because download was interrupted\n", c->cache_filename);
+          if (fclose(conn->file) != 0) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "error closing audio cache file\n");
+          }
+          conn->file = nullptr ;
+        }
+
+        if (c->cache_filename) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing audio cache file %s because download was interrupted\n", c->cache_filename);
+          if (unlink(c->cache_filename) != 0) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error removing audio cache file %s: %d:%s\n", 
+              c->cache_filename, errno, strerror(errno));
+          }
+          free(c->cache_filename);
+          c->cache_filename = nullptr ;
+        }
+      }
+    }
+    if (c->session_id) {
+      switch_core_session_t* session = switch_core_session_locate(c->session_id);
+      if (session) {
+        switch_channel_t *channel = switch_core_session_get_channel(session);
+        switch_core_session_rwunlock(session);
+        if (channel) {
+          switch_event_t *event;
+          if (switch_event_create(&event, SWITCH_EVENT_PLAYBACK_STOP) == SWITCH_STATUS_SUCCESS) {
+            switch_channel_event_set_data(channel, event);
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "Playback-File-Type", "tts_stream");
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_custom_response_code", std::to_string(c->response_code).c_str());
+            if (c->cache_filename && c->response_code == 200) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", c->cache_filename);
+            }
+            if (c->response_code != 200 && c->err_msg) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_error", c->err_msg);
+            }
+            switch_event_fire(&event);
+          }
+          else {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: failed to create event\n");
+          }
+        }
+        else {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: channel not found\n");
+        }
+      }
+    }
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+	switch_status_t custom_speech_close(custom_t* c) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "custom_speech_close\n") ;
+		return SWITCH_STATUS_SUCCESS;
+	}
+}

mod_custom_tts/custom_glue.h

@@ -0,0 +1,12 @@
+#ifndef __CUSTOM_GLUE_H__
+#define __CUSTOM_GLUE_H__
+
+switch_status_t custom_speech_load();
+switch_status_t custom_speech_open(custom_t* custom);
+switch_status_t custom_speech_feed_tts(custom_t* custom, char* text, switch_speech_flag_t *flags);
+switch_status_t custom_speech_read_tts(custom_t* custom, void *data, size_t *datalen, switch_speech_flag_t *flags);
+switch_status_t custom_speech_flush_tts(custom_t* custom);
+switch_status_t custom_speech_close(custom_t* custom);
+switch_status_t custom_speech_unload();
+
+#endif

mod_custom_tts/mod_custom_tts.c

@@ -0,0 +1,158 @@
+#include "mod_custom_tts.h"
+#include "custom_glue.h"
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_custom_tts_load);
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_custom_tts_shutdown);
+SWITCH_MODULE_DEFINITION(mod_custom_tts, mod_custom_tts_load, mod_custom_tts_shutdown, NULL);
+
+static void clearCustomVendor(custom_t* c, int freeAll) {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "clearCustomVendor\n");
+  if (c->auth_token) free(c->auth_token);
+  if (c->custom_tts_url) free(c->custom_tts_url);
+  if (c->language) free(c->language);
+  if (c->ct) free(c->ct);
+  if (c->err_msg) free(c->err_msg);
+  if (c->name_lookup_time_ms) free(c->name_lookup_time_ms);
+  if (c->connect_time_ms) free(c->connect_time_ms);
+  if (c->final_response_time_ms) free(c->final_response_time_ms);
+  if (c->cache_filename) free(c->cache_filename);
+  
+
+  c->auth_token = NULL;
+  c->custom_tts_url = NULL;
+  c->language = NULL;
+  c->ct = NULL;
+  c->err_msg = NULL;
+  c->name_lookup_time_ms = NULL;
+  c->connect_time_ms = NULL;
+  c->final_response_time_ms = NULL;
+  c->cache_filename = NULL;
+
+  if (freeAll) {
+    if (c->voice_name) free(c->voice_name);
+    if (c->session_id) free(c->session_id);
+    c->voice_name = NULL;
+    c->session_id = NULL;
+  }
+}
+
+static custom_t * createOrRetrievePrivateData(switch_speech_handle_t *sh) {
+  custom_t *c = (custom_t *) sh->private_info;  
+  if (!c) {
+    c = switch_core_alloc(sh->memory_pool, sizeof(*c));
+  	sh->private_info = c;
+    memset(c, 0, sizeof(*c));
+    switch_mutex_init(&c->mutex, SWITCH_MUTEX_NESTED, sh->memory_pool);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "allocated custom_t\n");
+  }
+  return c;
+}
+
+switch_status_t w_speech_open(switch_speech_handle_t *sh, const char *voice_name, int rate, int channels, switch_speech_flag_t *flags)
+{
+  custom_t *c = createOrRetrievePrivateData(sh);
+  c->voice_name = strdup(voice_name);
+  c->rate = rate;
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_speech_open voice: %s, rate %d, channels %d\n", voice_name, rate, channels);
+  return custom_speech_open(c);
+}
+
+static switch_status_t w_speech_close(switch_speech_handle_t *sh, switch_speech_flag_t *flags)
+{
+  switch_status_t rc;
+  custom_t *c = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "w_speech_close\n");
+
+  switch_mutex_destroy(c->mutex);
+
+  rc = custom_speech_close(c);
+  clearCustomVendor(c, 1);
+  return rc;
+}
+
+/**
+ * Freeswitch will call this function to feed us text to speak
+ */
+static switch_status_t w_speech_feed_tts(switch_speech_handle_t *sh, char *text, switch_speech_flag_t *flags)
+{
+  custom_t *c = createOrRetrievePrivateData(sh);
+  c->draining = 0;
+  c->reads = 0;
+  c->response_code = 0;
+  c->err_msg = NULL;
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "w_speech_feed_tts\n");
+
+  return custom_speech_feed_tts(c, text, flags);
+}
+
+/**
+ * Freeswitch calls periodically to get some rendered audio in L16 format. We can provide up to 8k of audio at a time.
+ */
+static switch_status_t w_speech_read_tts(switch_speech_handle_t *sh, void *data, size_t *datalen, switch_speech_flag_t *flags)
+{
+  custom_t *c = createOrRetrievePrivateData(sh);
+  return custom_speech_read_tts(c, data, datalen, flags);
+}
+
+/**
+ * This is called at the end, not sure exactly what we need to do here..
+ */
+static void w_speech_flush_tts(switch_speech_handle_t *sh)
+{
+  custom_t *c = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "w_speech_flush_tts\n");
+  custom_speech_flush_tts(c);
+
+  clearCustomVendor(c, 0);
+}
+
+static void w_text_param_tts(switch_speech_handle_t *sh, char *param, const char *val)
+{
+  custom_t *c = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "w_text_param_tts: %s=%s\n", param, val);
+  if (0 == strcmp(param, "auth_token")) {
+    if (c->auth_token) free(c->auth_token);
+    c->auth_token = strdup(val);
+  } else if (0 == strcmp(param, "custom_tts_url")) {
+    if (c->custom_tts_url) free(c->custom_tts_url);
+    c->custom_tts_url = strdup(val);
+  } else if (0 == strcmp(param, "language")) {
+    if (c->language) free(c->language);
+    c->language = strdup(val);
+  } else if (0 == strcmp(param, "session-uuid")) {
+    if (c->session_id) free(c->session_id);
+    c->session_id = strdup(val);
+  } else if (0 == strcmp(param, "write_cache_file") && switch_true(val)) {
+    c->cache_audio = 1;
+  }
+}
+static void w_numeric_param_tts(switch_speech_handle_t *sh, char *param, int val)
+{
+}
+static void w_float_param_tts(switch_speech_handle_t *sh, char *param, double val)
+{
+}
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_custom_tts_load)
+{
+  switch_speech_interface_t *speech_interface;
+
+  *module_interface = switch_loadable_module_create_module_interface(pool, modname);
+  speech_interface = switch_loadable_module_create_interface(*module_interface, SWITCH_SPEECH_INTERFACE);
+  speech_interface->interface_name = "custom";
+  speech_interface->speech_open = w_speech_open;
+  speech_interface->speech_close = w_speech_close;
+  speech_interface->speech_feed_tts = w_speech_feed_tts;
+  speech_interface->speech_read_tts = w_speech_read_tts;
+	speech_interface->speech_flush_tts = w_speech_flush_tts;
+	speech_interface->speech_text_param_tts = w_text_param_tts;
+	speech_interface->speech_numeric_param_tts = w_numeric_param_tts;
+	speech_interface->speech_float_param_tts = w_float_param_tts;
+  return custom_speech_load();
+}
+
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_custom_tts_shutdown)
+{
+  return custom_speech_unload();
+}

mod_custom_tts/mod_custom_tts.h

@@ -0,0 +1,31 @@
+#ifndef __MOD_CUSTOM_TTS_H__
+#define __MOD_CUSTOM_TTS_H__
+
+#include <switch.h>
+typedef struct custom_data {
+  char *voice_name;
+  char *auth_token;
+  char *custom_tts_url;
+  char *language;
+
+  /* result data */
+  long response_code;
+  char *ct;
+  char *name_lookup_time_ms;
+  char *connect_time_ms;
+  char *final_response_time_ms;
+  char *err_msg;
+  char *cache_filename;
+  char *session_id;
+
+  int rate;
+  int draining;
+  int reads;
+  int cache_audio;
+
+	void *conn;
+  void *circularBuffer;
+  switch_mutex_t *mutex;
+  FILE *file;
+} custom_t;
+#endif

mod_deepgram_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_deepgram_transcribe/README.md

@@ -25,10 +25,11 @@ Stop transcription on the channel.
 | variable | Description |
 | --- | ----------- |
 | DEEPGRAM_API_KEY | Deepgram API key used to authenticate |
-| DEEPGRAM_SPEECH_TIER | https://developers.deepgram.com/documentation/features/tier/ |
 | DEEPGRAM_SPEECH_CUSTOM_MODEL | custom model id |
 | DEEPGRAM_SPEECH_MODEL | https://developers.deepgram.com/documentation/features/model/ |
 | DEEPGRAM_SPEECH_MODEL_VERSION | https://developers.deepgram.com/documentation/features/version/ |
+| DEEPGRAM_SPEECH_ENABLE_SMART_FORMAT | https://developers.deepgram.com/docs/smart-format |
+| DEEPGRAM_SPEECH_ENABLE_FILLER_WORDS | https://developers.deepgram.com/docs/filler-words |
 | DEEPGRAM_SPEECH_ENABLE_AUTOMATIC_PUNCTUATION | https://developers.deepgram.com/documentation/features/punctuate/ |
 | DEEPGRAM_SPEECH_PROFANITY_FILTER | https://developers.deepgram.com/documentation/features/profanity-filter/ |
 | DEEPGRAM_SPEECH_REDACT | https://developers.deepgram.com/documentation/features/redact/ |
@@ -42,6 +43,7 @@ Stop transcription on the channel.
 | DEEPGRAM_SPEECH_REPLACE | https://developers.deepgram.com/documentation/features/replace/  |
 | DEEPGRAM_SPEECH_TAG | https://developers.deepgram.com/documentation/features/tag/ |
 | DEEPGRAM_SPEECH_ENDPOINTING  | https://developers.deepgram.com/documentation/features/endpointing/ |
+| DEEPGRAM_SPEECH_UTTERANCE_END_MS | https://developers.deepgram.com/docs/utterance-end |
 | DEEPGRAM_SPEECH_VAD_TURNOFF | https://developers.deepgram.com/documentation/features/voice-activity-detection/ |
 
 

mod_deepgram_transcribe/audio_pipe.cpp

@@ -176,7 +176,9 @@ int AudioPipe::lws_callback(struct lws *wsi,
           if (lws_is_final_fragment(wsi)) {
             if (nullptr != ap->m_recv_buf) {
               std::string msg((char *)ap->m_recv_buf, ap->m_recv_buf_ptr - ap->m_recv_buf);
-              ap->m_callback(ap->m_uuid.c_str(),  ap->m_bugname.c_str(), deepgram::AudioPipe::MESSAGE, msg.c_str(),  ap->isFinished());
+              if (!ap->m_silence_disconnect) {
+                ap->m_callback(ap->m_uuid.c_str(),  ap->m_bugname.c_str(), deepgram::AudioPipe::MESSAGE, msg.c_str(),  ap->isFinished());
+              }
               if (nullptr != ap->m_recv_buf) free(ap->m_recv_buf);
             }
             ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
@@ -451,7 +453,7 @@ AudioPipe::AudioPipe(const char* uuid, const char* bugname, const char* host, un
   m_uuid(uuid), m_host(host), m_port(port), m_path(path), m_finished(false), m_bugname(bugname),
   m_audio_buffer_min_freespace(minFreespace), m_audio_buffer_max_len(bufLen), m_gracefulShutdown(false),
   m_audio_buffer_write_offset(LWS_PRE), m_recv_buf(nullptr), m_recv_buf_ptr(nullptr), m_useTls(useTls),
-  m_state(LWS_CLIENT_IDLE), m_wsi(nullptr), m_vhd(nullptr), m_callback(callback) {
+  m_state(LWS_CLIENT_IDLE), m_wsi(nullptr), m_vhd(nullptr), m_callback(callback), m_silence_disconnect(false) {
 
   if (apiKey) m_apiKey = apiKey;
   else m_apiKey = "";
@@ -516,6 +518,11 @@ void AudioPipe::finish() {
   bufferForSending("{\"type\": \"CloseStream\"}");
 }
 
+void AudioPipe::finish_in_silence() {
+  m_silence_disconnect = true;
+  finish();
+}
+
 void AudioPipe::waitForClose() {
   std::shared_future<void> sf(m_promise.get_future());
   sf.wait();

mod_deepgram_transcribe/audio_pipe.hpp

@@ -76,6 +76,7 @@ namespace deepgram {
 
     void close() ;
     void finish();
+    void finish_in_silence();
     void waitForClose();
     void setClosed() { m_promise.set_value(); }
     bool isFinished() { return m_finished;}
@@ -138,6 +139,7 @@ namespace deepgram {
     std::string m_bugname;
     std::promise<void> m_promise;
     bool m_useTls;
+    bool m_silence_disconnect;
   };
 
 } // namespace deepgram

mod_deepgram_transcribe/dg_transcribe_glue.cpp

@@ -22,6 +22,7 @@
 
 #define RTP_PACKETIZATION_PERIOD 20
 #define FRAME_SIZE_8000  320 /*which means each 20ms frame as 320 bytes at 8 khz (1 channel only)*/
+#define DEEPGRAM_KEEP_ALIVE_INTERVAL_SECOND 8
 
 namespace {
   static bool hasDefaultCredentials = false;
@@ -86,13 +87,17 @@ namespace {
   static const char* emptyTranscript = 
     "\"is_final\":false,\"speech_final\":false,\"channel\":{\"alternatives\":[{\"transcript\":\"\",\"confidence\":0.0,\"words\":[]}]}";
 
-  static void reaper(private_t *tech_pvt) {
+  static void reaper(private_t *tech_pvt, bool silence_disconnect) {
     std::shared_ptr<deepgram::AudioPipe> pAp;
     pAp.reset((deepgram::AudioPipe *)tech_pvt->pAudioPipe);
     tech_pvt->pAudioPipe = nullptr;
 
-    std::thread t([pAp, tech_pvt]{
-      pAp->finish();
+    std::thread t([pAp, tech_pvt, silence_disconnect]{
+      if (silence_disconnect) {
+        pAp->finish_in_silence();
+      } else {
+        pAp->finish();
+      }
       pAp->waitForClose();
       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "%s (%u) got remote close\n", tech_pvt->sessionId, tech_pvt->id);
     });
@@ -112,6 +117,8 @@ namespace {
           tech_pvt->resampler = NULL;
       }
 
+      // NB: do not destroy the mutex here, that is caller responsibility
+
       /*
       if (tech_pvt->vad) {
         switch_vad_destroy(&tech_pvt->vad);
@@ -165,7 +172,7 @@ namespace {
     else if (customModel) oss << "&model=" << customModel;
 
     if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_MODEL_VERSION")) {
-     oss <<  "&version";
+     oss <<  "&version=";
      oss <<  var;
     }
     oss <<  "&language=";
@@ -184,6 +191,9 @@ namespace {
       * 
       */
     }
+    if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_ENABLE_FILLER_WORDS")) {
+     oss <<  "&filler_words=true";
+    }
     if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_ENABLE_AUTOMATIC_PUNCTUATION")) {
      oss <<  "&punctuate=true";
     }
@@ -292,9 +302,20 @@ namespace {
             break;
             case deepgram::AudioPipe::CONNECTION_DROPPED:
               // first thing: we can no longer access the AudioPipe
-              tech_pvt->pAudioPipe = nullptr;
-              tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_DISCONNECT, NULL, tech_pvt->bugname, finished);
-              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection (%s) dropped from far end\n", tech_pvt->bugname);
+
+              /**
+               * this is a bit tricky.  If we just closed a previos connection it may be returning final transcripts
+               * and then a close event here as it is shutting down (in the reaper thread above).
+               * In this scenario, the fact that the connection is dropped is not significant.
+               */
+              if (finished) {
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "old connection (%s) gracefully closed by Deepgram\n", tech_pvt->bugname);
+              }
+              else {
+                tech_pvt->pAudioPipe = nullptr;
+                tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_DISCONNECT, NULL, tech_pvt->bugname, finished);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection (%s) dropped from far end\n", tech_pvt->bugname);
+              }
             break;
             case deepgram::AudioPipe::CONNECTION_CLOSED_GRACEFULLY:
               // first thing: we can no longer access the AudioPipe
@@ -326,60 +347,82 @@ namespace {
 
     int err;
     int useTls = true;
+    std::string host;
+    int port = 443;
+    size_t buflen = LWS_PRE + (FRAME_SIZE_8000 * desiredSampling / 8000 * channels * 1000 / RTP_PACKETIZATION_PERIOD * nAudioBufferSecs);
+
+    std::ostringstream configuration_stream;
     switch_codec_implementation_t read_impl;
     switch_channel_t *channel = switch_core_session_get_channel(session);
 
     switch_core_session_get_read_impl(session, &read_impl);
   
-    memset(tech_pvt, 0, sizeof(private_t));
-  
     std::string path;
     constructPath(session, path, desiredSampling, channels, lang, interim);
     switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "path: %s\n", path.c_str());
 
-    strncpy(tech_pvt->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
-
     const char* endpoint = switch_channel_get_variable(channel, "DEEPGRAM_URI");
     if (endpoint != nullptr) {
       std::string ep(endpoint);
-
       useTls = switch_true(switch_channel_get_variable(channel, "DEEPGRAM_USE_TLS"));
 
       size_t pos = ep.find(':');
+      host = ep;
       if (pos != std::string::npos) {
-        std::string host = ep.substr(0, pos);
-        std::string port = ep.substr(pos + 1);
-        strncpy(tech_pvt->host, host.c_str(), MAX_WS_URL_LEN);
-        tech_pvt->port = ::atoi(port.c_str());
-      }
-      else {
-        strncpy(tech_pvt->host, ep.c_str(), MAX_WS_URL_LEN);
-        tech_pvt->port = 443;
+        host = ep.substr(0, pos);
+        std::string strPort = ep.substr(pos + 1);
+        port = ::atoi(strPort.c_str());
       }
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, 
-        "connecting to deepgram on-prem %s port %d, using tls? (%s)\n", tech_pvt->host, tech_pvt->port, useTls ? "yes" : "no");
-    }
-    else {
-      strncpy(tech_pvt->host, "api.deepgram.com", MAX_WS_URL_LEN);
-      tech_pvt->port = 443;
+        "connecting to deepgram on-prem %s port %d, using tls? (%s)\n", host.c_str(), port, useTls ? "yes" : "no");
+    } else {
+      host = "api.deepgram.com";
     }
 
-    strncpy(tech_pvt->path, path.c_str(), MAX_PATH_LEN);    
+    const char* apiKey = switch_channel_get_variable(channel, "DEEPGRAM_API_KEY");
+    if (!apiKey && defaultApiKey) {
+      apiKey = defaultApiKey;
+    } else if (!apiKey && endpoint == nullptr) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "no deepgram api key provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    configuration_stream <<
+      host << ":" <<
+      port << ";" <<
+      path << ";" <<
+      buflen << ";" <<
+      read_impl.decoded_bytes_per_packet << ";" <<
+      apiKey << ";" <<
+      useTls;
+
+    if (tech_pvt->pAudioPipe) {
+      // stop sending keep alive
+      tech_pvt->is_keep_alive = 0;
+      if (0 != strcmp(tech_pvt->configuration, configuration_stream.str().c_str())) {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "fork_data_init: stop existing deepgram connection, old configuration %s, new configuration %s\n",
+          tech_pvt->configuration, configuration_stream.str().c_str());
+        reaper(tech_pvt, true);
+      } else {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "fork_data_init: enable existing deepgram connection\n");
+        return SWITCH_STATUS_SUCCESS;
+      }
+    } else {
+      memset(tech_pvt, 0, sizeof(private_t));
+    }
+
+    strncpy(tech_pvt->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
+    strncpy(tech_pvt->host, host.c_str(), MAX_WS_URL_LEN);
+    tech_pvt->port = port;
+    strncpy(tech_pvt->path, path.c_str(), MAX_PATH_LEN);   
+    strncpy(tech_pvt->configuration, configuration_stream.str().c_str(), MAX_PATH_LEN) ;
     tech_pvt->sampling = desiredSampling;
     tech_pvt->responseHandler = responseHandler;
     tech_pvt->channels = channels;
     tech_pvt->id = ++idxCallCount;
     tech_pvt->buffer_overrun_notified = 0;
+    tech_pvt->is_keep_alive = 0;
     strncpy(tech_pvt->bugname, bugname, MAX_BUG_LEN);
-    
-    size_t buflen = LWS_PRE + (FRAME_SIZE_8000 * desiredSampling / 8000 * channels * 1000 / RTP_PACKETIZATION_PERIOD * nAudioBufferSecs);
-
-    const char* apiKey = switch_channel_get_variable(channel, "DEEPGRAM_API_KEY");
-    if (!apiKey && defaultApiKey) apiKey = defaultApiKey;
-    else if (!apiKey && endpoint == nullptr) {
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "no deepgram api key provided\n");
-      return SWITCH_STATUS_FALSE;
-    }
 
     deepgram::AudioPipe* ap = new deepgram::AudioPipe(tech_pvt->sessionId, bugname, tech_pvt->host, tech_pvt->port, tech_pvt->path, 
       buflen, read_impl.decoded_bytes_per_packet, apiKey, useTls, eventCallback);
@@ -390,18 +433,26 @@ namespace {
 
     tech_pvt->pAudioPipe = static_cast<void *>(ap);
 
-    switch_mutex_init(&tech_pvt->mutex, SWITCH_MUTEX_NESTED, switch_core_session_get_pool(session));
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connecting now\n");
+    ap->connect();
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection in progress\n");
 
-    if (desiredSampling != sampling) {
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) resampling from %u to %u\n", tech_pvt->id, sampling, desiredSampling);
-      tech_pvt->resampler = speex_resampler_init(channels, sampling, desiredSampling, SWITCH_RESAMPLE_QUALITY, &err);
-      if (0 != err) {
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing resampler: %s.\n", speex_resampler_strerror(err));
-        return SWITCH_STATUS_FALSE;
-      }
+    if (!tech_pvt->mutex) {
+      switch_mutex_init(&tech_pvt->mutex, SWITCH_MUTEX_NESTED, switch_core_session_get_pool(session));
     }
-    else {
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) no resampling needed for this call\n", tech_pvt->id);
+    
+    if (!tech_pvt->resampler) {
+      if (desiredSampling != sampling) {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) resampling from %u to %u\n", tech_pvt->id, sampling, desiredSampling);
+        tech_pvt->resampler = speex_resampler_init(channels, sampling, desiredSampling, SWITCH_RESAMPLE_QUALITY, &err);
+        if (0 != err) {
+          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error initializing resampler: %s.\n", speex_resampler_strerror(err));
+          return SWITCH_STATUS_FALSE;
+        }
+      }
+      else {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) no resampling needed for this call\n", tech_pvt->id);
+      }
     }
 
     switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) fork_data_init\n", tech_pvt->id);
@@ -458,11 +509,22 @@ extern "C" {
   switch_status_t dg_transcribe_session_init(switch_core_session_t *session, 
     responseHandler_t responseHandler, uint32_t samples_per_second, uint32_t channels, 
     char* lang, int interim, char* bugname, void **ppUserData)
-  {    	
+  {
     int err;
+    switch_channel_t *channel = switch_core_session_get_channel(session);
+    switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, bugname);
+    private_t* tech_pvt;
+    if (bug) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "reuse existing kep alive deepgram connection\n");
+      tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
+    } else {
+      tech_pvt = (private_t *) switch_core_session_alloc(session, sizeof(private_t));
+      tech_pvt->pAudioPipe = NULL;
+      tech_pvt->is_keep_alive = 0;
+      tech_pvt->mutex = NULL;
+      tech_pvt->resampler = NULL;
+    }
 
-    // allocate per-session data structure
-    private_t* tech_pvt = (private_t *) switch_core_session_alloc(session, sizeof(private_t));
     if (!tech_pvt) {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "error allocating memory!\n");
       return SWITCH_STATUS_FALSE;
@@ -475,16 +537,13 @@ extern "C" {
 
     *ppUserData = tech_pvt;
 
-    deepgram::AudioPipe *pAudioPipe = static_cast<deepgram::AudioPipe *>(tech_pvt->pAudioPipe);
-    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connecting now\n");
-    pAudioPipe->connect();
-    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection in progress\n");
     return SWITCH_STATUS_SUCCESS;
   }
 
 	switch_status_t dg_transcribe_session_stop(switch_core_session_t *session,int channelIsClosing, char* bugname) {
     switch_channel_t *channel = switch_core_session_get_channel(session);
     switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, bugname);
+    const bool use_single_connection = switch_true(std::getenv("DEEPGRAM_SPEECH_USE_SINGLE_CONNECTION"));
     if (!bug) {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "dg_transcribe_session_stop: no bug - websocket conection already closed\n");
       return SWITCH_STATUS_FALSE;
@@ -495,6 +554,12 @@ extern "C" {
     switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) dg_transcribe_session_stop\n", id);
 
     if (!tech_pvt) return SWITCH_STATUS_FALSE;
+    if (use_single_connection && !channelIsClosing) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "dg_transcribe_session_stop: call is running, use_single_connection is true, keep alive is activated\n", id);
+      tech_pvt->is_keep_alive = 1;
+      tech_pvt->frame_count = 0;
+      return SWITCH_STATUS_SUCCESS;
+    }
       
     // close connection and get final responses
     switch_mutex_lock(tech_pvt->mutex);
@@ -502,7 +567,7 @@ extern "C" {
     if (!channelIsClosing) switch_core_media_bug_remove(session, &bug);
 
     deepgram::AudioPipe *pAudioPipe = static_cast<deepgram::AudioPipe *>(tech_pvt->pAudioPipe);
-    if (pAudioPipe) reaper(tech_pvt);
+    if (pAudioPipe) reaper(tech_pvt, false);
     destroy_tech_pvt(tech_pvt);
     switch_mutex_unlock(tech_pvt->mutex);
     switch_mutex_destroy(tech_pvt->mutex);
@@ -516,9 +581,31 @@ extern "C" {
     size_t inuse = 0;
     bool dirty = false;
     char *p = (char *) "{\"msg\": \"buffer overrun\"}";
+    char *keep_alive = (char *) "{\"type\": \"KeepAlive\"}";
 
     if (!tech_pvt) return SWITCH_TRUE;
     
+
+    // Keep sending keep alive if there is no transcribe activity
+    if (tech_pvt->is_keep_alive && tech_pvt->pAudioPipe) {
+      deepgram::AudioPipe *pAudioPipe = static_cast<deepgram::AudioPipe *>(tech_pvt->pAudioPipe);
+      if (++tech_pvt->frame_count * 20 /*ms*/ / 1000 >= DEEPGRAM_KEEP_ALIVE_INTERVAL_SECOND) {
+        tech_pvt->frame_count = 0;
+        pAudioPipe->bufferForSending(keep_alive);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "dg_transcribe_frame: sending %s to deepgram\n", keep_alive);
+      }
+      // remove media bug buffered data
+      while (true) {
+        unsigned char data[SWITCH_RECOMMENDED_BUFFER_SIZE] = {0};
+        switch_frame_t frame = { 0 };
+        frame.data = data;
+        frame.buflen = SWITCH_RECOMMENDED_BUFFER_SIZE;
+        switch_status_t rv = switch_core_media_bug_read(bug, &frame, SWITCH_TRUE);
+        if (rv != SWITCH_STATUS_SUCCESS) break;
+      }
+      return SWITCH_TRUE;
+    }
+    
     if (switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
       if (!tech_pvt->pAudioPipe) {
         switch_mutex_unlock(tech_pvt->mutex);
@@ -529,7 +616,6 @@ extern "C" {
         switch_mutex_unlock(tech_pvt->mutex);
         return SWITCH_TRUE;
       }
-
       pAudioPipe->lockAudioBuffer();
       size_t available = pAudioPipe->binarySpaceAvailable();
       if (NULL == tech_pvt->resampler) {
@@ -571,7 +657,6 @@ extern "C" {
           if (frame.datalen) {
             spx_uint32_t out_len = available >> 1;  // space for samples which are 2 bytes
             spx_uint32_t in_len = frame.samples;
-
             speex_resampler_process_interleaved_int(tech_pvt->resampler, 
               (const spx_int16_t *) frame.data, 
               (spx_uint32_t *) &in_len, 

mod_deepgram_transcribe/mod_deepgram_transcribe.c

@@ -5,6 +5,7 @@
  */
 #include "mod_deepgram_transcribe.h"
 #include "dg_transcribe_glue.h"
+#include <stdio.h>
 
 /* Prototypes */
 SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_deepgram_transcribe_shutdown);
@@ -72,8 +73,10 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	switch_codec_implementation_t read_impl = { 0 };
 	void *pUserData;
 	uint32_t samples_per_second;
+	int use_single_connection = switch_true(getenv("DEEPGRAM_SPEECH_USE_SINGLE_CONNECTION"));
+	bug = switch_channel_get_private(channel, bugname);
 
-	if (switch_channel_get_private(channel, bugname)) {
+	if (bug && !use_single_connection) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing bug from previous transcribe\n");
 		do_stop(session, bugname);
 	}
@@ -90,11 +93,13 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing dg speech session.\n");
 		return SWITCH_STATUS_FALSE;
 	}
-	if ((status = switch_core_media_bug_add(session, "dg_transcribe", NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
-		return status;
+	if (!bug || !use_single_connection) {
+		if ((status = switch_core_media_bug_add(session, "dg_transcribe", NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
+			return status;
+		}
+		switch_channel_set_private(channel, bugname, bug);
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for dg transcribe\n");
 	}
-  switch_channel_set_private(channel, bugname, bug);
-	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for dg transcribe\n");
 
 	return SWITCH_STATUS_SUCCESS;
 }

mod_deepgram_transcribe/mod_deepgram_transcribe.h

@@ -39,6 +39,9 @@ struct private_data {
   unsigned int id;
   int buffer_overrun_notified:1;
   int is_finished:1;
+  int is_keep_alive;
+  int frame_count;
+  char configuration[MAX_PATH_LEN];
 };
 
 typedef struct private_data private_t;

mod_deepgram_tts/Makefile.am

@@ -0,0 +1,8 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_deepgram_tts
+
+mod_LTLIBRARIES = mod_deepgram_tts.la
+mod_deepgram_tts_la_SOURCES  = mod_deepgram_tts.c deepgram_glue.cpp
+mod_deepgram_tts_la_CFLAGS   = $(AM_CFLAGS)
+mod_deepgram_tts_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_deepgram_tts_la_LDFLAGS  = -avoid-version -module -no-undefined -shared -lstdc++ -lboost_system -lboost_thread

mod_deepgram_tts/deepgram_glue.cpp

@@ -0,0 +1,976 @@
+#include "mod_deepgram_tts.h"
+#include <switch.h>
+#include <switch_json.h>
+#include <curl/curl.h>
+#include <cstdlib>
+
+#include <boost/circular_buffer.hpp>
+#include <boost/thread.hpp>
+#include <boost/asio.hpp>
+#include <boost/asio/ssl.hpp>
+#include <boost/pool/object_pool.hpp>
+#include <boost/bind/bind.hpp>
+#include <boost/tokenizer.hpp>
+#include <boost/foreach.hpp>
+#include <boost/asio.hpp>
+#include <boost/assign/list_of.hpp>
+#include <boost/algorithm/string.hpp>
+
+#include <speex/speex_resampler.h>
+
+#define BUFFER_GROW_SIZE (80000)
+
+typedef boost::circular_buffer<uint16_t> CircularBuffer_t;
+/* Global information, common to all connections */
+typedef struct
+{
+  CURLM *multi;
+  int still_running;
+} GlobalInfo_t;
+static GlobalInfo_t global;
+
+/* Information associated with a specific easy handle */
+typedef struct
+{
+  CURL *easy;
+  deepgram_t* deepgram;
+  char* body;
+  struct curl_slist *hdr_list;
+  GlobalInfo_t *global;
+  char error[CURL_ERROR_SIZE];
+  FILE* file;
+  std::chrono::time_point<std::chrono::high_resolution_clock> startTime;
+  bool flushed;
+
+  bool has_last_byte;
+  uint8_t last_byte;
+} ConnInfo_t;
+
+
+static boost::object_pool<ConnInfo_t> pool ;
+static std::map<curl_socket_t, boost::asio::ip::tcp::socket *> socket_map;
+static boost::asio::io_service io_service;
+static boost::asio::deadline_timer timer(io_service);
+static std::string fullDirPath;
+static std::thread worker_thread;
+
+std::string secondsToMillisecondsString(double seconds) {
+    // Convert to milliseconds
+    double milliseconds = seconds * 1000.0;
+
+    // Truncate to remove fractional part
+    long milliseconds_long = static_cast<long>(milliseconds);
+
+    // Convert to string
+    return std::to_string(milliseconds_long);
+}
+
+static std::string getEnvVar(const std::string& varName) {
+    const char* val = std::getenv(varName.c_str());
+    return val ? std::string(val) : "";
+}
+
+static long getConnectionTimeout() {
+    std::string connectTimeoutStr = getEnvVar("DEEPGRAM_TTS_CURL_CONNECT_TIMEOUT");
+
+    if (connectTimeoutStr.empty()) {
+        connectTimeoutStr = getEnvVar("TTS_CURL_CONNECT_TIMEOUT");
+    }
+
+    long connectTimeout = 10L;
+
+    if (!connectTimeoutStr.empty()) {
+        connectTimeout = std::stol(connectTimeoutStr);
+    }
+
+    return connectTimeout;
+}
+
+static CURL* createEasyHandle(void) {
+  CURL* easy = curl_easy_init();
+  if(!easy) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "curl_easy_init() failed!\n");
+    return nullptr ;
+  }  
+
+  curl_easy_setopt(easy, CURLOPT_FOLLOWLOCATION, 1L);
+  curl_easy_setopt(easy, CURLOPT_USERAGENT, "jambonz/0.8.5");
+
+  // set connect timeout to 3 seconds and total timeout to 109 seconds
+  curl_easy_setopt(easy, CURLOPT_CONNECTTIMEOUT_MS, 3000L);
+  curl_easy_setopt(easy, CURLOPT_TIMEOUT, getConnectionTimeout());
+
+  return easy ;    
+}
+
+static void cleanupConn(ConnInfo_t *conn) {
+  auto d = conn->deepgram;
+
+  if( conn->hdr_list ) {
+    curl_slist_free_all(conn->hdr_list);
+    conn->hdr_list = nullptr ;
+  }
+  curl_easy_cleanup(conn->easy);
+
+  if (conn->file) {
+    if (fclose(conn->file) != 0) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error closing audio cache file\n");
+    }
+    conn->file = nullptr ;
+  }
+
+  d->conn = nullptr ;
+  d->draining = 1;
+
+  memset(conn, 0, sizeof(ConnInfo_t));
+  pool.destroy(conn) ;
+}
+
+/* Check for completed transfers, and remove their easy handles */
+void check_multi_info(GlobalInfo_t *g) {
+  CURLMsg *msg;
+  int msgs_left;
+  ConnInfo_t *conn;
+  CURL *easy;
+  CURLcode res;
+  
+  while((msg = curl_multi_info_read(g->multi, &msgs_left))) {
+    if(msg->msg == CURLMSG_DONE) {
+      long response_code;
+      double namelookup=0, connect=0, total=0 ;
+      char *ct = NULL ;
+
+      easy = msg->easy_handle;
+      res = msg->data.result;
+      curl_easy_getinfo(easy, CURLINFO_PRIVATE, &conn);
+      curl_easy_getinfo(easy, CURLINFO_RESPONSE_CODE, &response_code);
+      curl_easy_getinfo(easy, CURLINFO_CONTENT_TYPE, &ct);
+
+      curl_easy_getinfo(easy, CURLINFO_NAMELOOKUP_TIME, &namelookup);
+      curl_easy_getinfo(easy, CURLINFO_CONNECT_TIME, &connect);
+      curl_easy_getinfo(easy, CURLINFO_TOTAL_TIME, &total);
+
+      auto d = conn->deepgram;
+      d->response_code = response_code;
+      if (ct) d->ct = strdup(ct);
+
+      std::string name_lookup_ms = secondsToMillisecondsString(namelookup);
+      std::string connect_ms = secondsToMillisecondsString(connect);
+      std::string final_response_time_ms = secondsToMillisecondsString(total);
+
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, 
+        "mod_deepgram_tts: response: %ld, content-type %s,"
+        "dns(ms): %"  CURL_FORMAT_CURL_OFF_T ".%06ld, "
+        "connect(ms): %"  CURL_FORMAT_CURL_OFF_T ".%06ld, "
+        "total(ms): %"  CURL_FORMAT_CURL_OFF_T ".%06ld\n",
+        response_code, ct,
+        (long)(namelookup), (long)(fmod(namelookup, 1.0) * 1000000),
+        (long)(connect), (long)(fmod(connect, 1.0) * 1000000),
+        (long)(total), (long)(fmod(total, 1.0) * 1000000));
+
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "name lookup time: %s\n", name_lookup_ms.c_str());
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "connect time: %s\n", connect_ms.c_str());
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "final response time: %s\n", final_response_time_ms.c_str());
+
+      d->name_lookup_time_ms = strdup(name_lookup_ms.c_str());
+      d->connect_time_ms = strdup(connect_ms.c_str());
+      d->final_response_time_ms = strdup(final_response_time_ms.c_str());
+
+      curl_multi_remove_handle(g->multi, easy);
+      cleanupConn(conn);
+    }
+  }
+}
+
+int mcode_test(const char *where, CURLMcode code) {
+  if(CURLM_OK != code) {
+    const char *s;
+    switch(code) {
+    case CURLM_CALL_MULTI_PERFORM:
+      s = "CURLM_CALL_MULTI_PERFORM";
+      break;
+    case CURLM_BAD_HANDLE:
+      s = "CURLM_BAD_HANDLE";
+      break;
+    case CURLM_BAD_EASY_HANDLE:
+      s = "CURLM_BAD_EASY_HANDLE";
+      break;
+    case CURLM_OUT_OF_MEMORY:
+      s = "CURLM_OUT_OF_MEMORY";
+      break;
+    case CURLM_INTERNAL_ERROR:
+      s = "CURLM_INTERNAL_ERROR";
+      break;
+    case CURLM_UNKNOWN_OPTION:
+      s = "CURLM_UNKNOWN_OPTION";
+      break;
+    case CURLM_LAST:
+      s = "CURLM_LAST";
+      break;
+    default:
+      s = "CURLM_unknown";
+      break;
+    case CURLM_BAD_SOCKET:
+      s = "CURLM_BAD_SOCKET";
+      break;
+    }
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mcode_test ERROR: %s returns %s:%d\n", where, s, code);
+
+    return -1;
+  }
+  return 0 ;
+}
+
+static void remsock(int *f, GlobalInfo_t *g) {
+  if(f) {
+    free(f);
+    f = NULL;
+  }
+}
+
+/* Called by asio when there is an action on a socket */
+static void event_cb(GlobalInfo_t *g, curl_socket_t s, int action, const boost::system::error_code & error, int *fdp) {
+  int f = *fdp;
+
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "event_cb socket %#X has action %d\n", s, action) ; 
+
+  // Socket already POOL REMOVED.
+  if (f == CURL_POLL_REMOVE) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "event_cb socket %#X removed\n", s); 
+    remsock(fdp, g);
+    return;
+  }
+
+  if(socket_map.find(s) == socket_map.end()) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "event_cb: socket  %#X already closed\n, s");
+    return;
+  }
+
+  /* make sure the event matches what are wanted */
+  if(f == action || f == CURL_POLL_INOUT) {
+    if(error) {
+      action = CURL_CSELECT_ERR;
+    }
+    CURLMcode rc = curl_multi_socket_action(g->multi, s, action, &g->still_running);
+
+    mcode_test("event_cb: curl_multi_socket_action", rc);
+    check_multi_info(g);
+
+    if(g->still_running <= 0) {
+      timer.cancel();
+    }
+
+    /* keep on watching.
+      * the socket may have been closed and/or fdp may have been changed
+      * in curl_multi_socket_action(), so check them both */
+    if(!error && socket_map.find(s) != socket_map.end() &&
+        (f == action || f == CURL_POLL_INOUT)) {
+      boost::asio::ip::tcp::socket *tcp_socket = socket_map.find(s)->second;
+
+      if(action == CURL_POLL_IN) {
+        tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                                action, boost::placeholders::_1, fdp));
+      }
+      if(action == CURL_POLL_OUT) {
+        tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                      boost::bind(&event_cb, g, s,
+                                                  action, boost::placeholders::_1, fdp));
+      } 
+    }
+  }
+}
+
+/* socket functions */
+static void setsock(int *fdp, curl_socket_t s, CURL *e, int act, int oldact, GlobalInfo_t *g) {
+  std::map<curl_socket_t, boost::asio::ip::tcp::socket *>::iterator it = socket_map.find(s);
+
+  if(it == socket_map.end()) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "setsock: socket  %#X not found\n, s");
+    return;
+  }
+
+  boost::asio::ip::tcp::socket * tcp_socket = it->second;
+
+  *fdp = act;
+
+  if(act == CURL_POLL_IN) {
+    if(oldact != CURL_POLL_IN && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                  boost::bind(&event_cb, g, s,
+                                  CURL_POLL_IN, boost::placeholders::_1, fdp));
+    }
+  }
+  else if(act == CURL_POLL_OUT) {
+    if(oldact != CURL_POLL_OUT && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                    CURL_POLL_OUT, boost::placeholders::_1, fdp));
+    }
+  }
+  else if(act == CURL_POLL_INOUT) {
+    if(oldact != CURL_POLL_IN && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                  boost::bind(&event_cb, g, s,
+                                  CURL_POLL_IN, boost::placeholders::_1, fdp));
+    }
+    if(oldact != CURL_POLL_OUT && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                    CURL_POLL_OUT, boost::placeholders::_1, fdp));
+    }
+  }
+}
+
+static void addsock(curl_socket_t s, CURL *easy, int action, GlobalInfo_t *g) {
+  /* fdp is used to store current action */
+  int *fdp = (int *) calloc(sizeof(int), 1);
+
+  setsock(fdp, s, easy, action, 0, g);
+  curl_multi_assign(g->multi, s, fdp);
+}
+
+static int sock_cb(CURL *e, curl_socket_t s, int what, void *cbp, void *sockp) {
+  GlobalInfo_t *g = &global;
+
+  int *actionp = (int *) sockp;
+  static const char *whatstr[] = { "none", "IN", "OUT", "INOUT", "REMOVE"};
+
+  if(what == CURL_POLL_REMOVE) {
+    *actionp = what;
+  }
+  else {
+    if(!actionp) {
+      addsock(s, e, what, g);
+    }
+    else {
+      setsock(actionp, s, e, what, *actionp, g);
+    }
+  }
+  return 0;  
+}
+
+static void threadFunc() {      
+  /* to make sure the event loop doesn't terminate when there is no work to do */
+  io_service.reset() ;
+  boost::asio::io_service::work work(io_service);
+  
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_deepgram_tts threadFunc - starting\n");
+
+  for(;;) {
+      
+    try {
+      io_service.run() ;
+      break ;
+    }
+    catch( std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mod_deepgram_tts threadFunc - Error: %s\n", e.what());
+    }
+  }
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_deepgram_tts threadFunc - ending\n");
+}
+
+
+/* Called by asio when our timeout expires */
+static void timer_cb(const boost::system::error_code & error, GlobalInfo_t *g)
+{
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "timer_cb\n");
+
+  if(!error) {
+    CURLMcode rc = curl_multi_socket_action(g->multi, CURL_SOCKET_TIMEOUT, 0, &g->still_running);
+    mcode_test("timer_cb: curl_multi_socket_action", rc);
+    check_multi_info(g);
+  }
+}
+
+int multi_timer_cb(CURLM *multi, long timeout_ms, GlobalInfo_t *g) {
+
+  /* cancel running timer */
+  timer.cancel();
+
+  if(timeout_ms >= 0) {
+    // from libcurl 7.88.1-10+deb12u4 does not allow call curl_multi_socket_action or curl_multi_perform in curl_multi callback directly
+    timer.expires_from_now(boost::posix_time::millisec(timeout_ms ? timeout_ms : 1));
+    timer.async_wait(boost::bind(&timer_cb, boost::placeholders::_1, g));
+  }
+
+  return 0;
+}
+
+/* CURLOPT_WRITEFUNCTION */
+static size_t write_cb(void *ptr, size_t size, size_t nmemb, ConnInfo_t *conn) {
+  bool fireEvent = false;
+  uint8_t *data = (uint8_t *) ptr;
+  size_t bytes_received = size * nmemb;
+  size_t total_bytes_to_process;
+  auto d = conn->deepgram;
+  CircularBuffer_t *cBuffer = (CircularBuffer_t *) d->circularBuffer;
+  
+  if (conn->flushed || cBuffer == nullptr) {
+    /* this will abort the transfer */
+    return 0;
+  }
+  // Buffer to hold combined data if there is unprocessed byte from the last call.
+  std::unique_ptr<uint8_t[]> combinedData;
+
+  if (conn->has_last_byte) {
+    conn->has_last_byte = false;  // We'll handle the last_byte now, so toggle the flag off
+
+    // Allocate memory for the new data array
+    combinedData.reset(new uint8_t[bytes_received + 1]);
+
+    // Prepend the last byte from previous call
+    combinedData[0] = conn->last_byte;
+
+    // Copy the new data following the prepended byte
+    memcpy(combinedData.get() + 1, data, bytes_received);
+
+    // Point our data pointer to the new array
+    data = combinedData.get();
+
+    total_bytes_to_process = bytes_received + 1;
+  } else {
+    total_bytes_to_process = bytes_received;
+  }
+
+  // If we now have an odd total, save the last byte for next time
+  if ((total_bytes_to_process % sizeof(int16_t)) != 0) {
+    conn->last_byte = data[total_bytes_to_process - 1];
+    conn->has_last_byte = true;
+    total_bytes_to_process--;
+  }
+
+  int16_t* inputData = reinterpret_cast<int16_t*>(data);
+  if (0 == d->reads++) {
+    fireEvent = true;
+    // Deepgram return PCM linear16 WAV file which contains 44 bytes headers, remove that.
+    inputData += 22;
+    total_bytes_to_process -= 44;
+  }
+  size_t numSamples = total_bytes_to_process / sizeof(int16_t);
+  {
+    switch_mutex_lock(d->mutex);
+
+    if (d->response_code > 0 && d->response_code != 200) {
+      std::string body((char *) ptr, bytes_received);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: received body %s\n", body.c_str());
+      d->err_msg = strdup(body.c_str());
+      switch_mutex_unlock(d->mutex);
+      return 0;
+    }
+
+    /* cache file will stay in the mp3 format for size (smaller) and simplicity */
+    if (conn->file) fwrite(inputData, sizeof(int16_t), numSamples, conn->file);
+
+    // Resize the buffer if necessary
+    if (cBuffer->capacity() - cBuffer->size() < numSamples) {
+      //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "write_cb growing buffer\n"); 
+
+      //TODO: if buffer exceeds some max size, return CURL_WRITEFUNC_ERROR to abort the transfer
+      cBuffer->set_capacity(cBuffer->size() + std::max(numSamples, (size_t)BUFFER_GROW_SIZE));
+    }
+    
+    /* Push the data into the buffer */
+    cBuffer->insert(cBuffer->end(), inputData, inputData + numSamples);
+
+    switch_mutex_unlock(d->mutex);
+  }
+  if (fireEvent && d->session_id) {
+    auto endTime = std::chrono::high_resolution_clock::now();
+    auto duration = std::chrono::duration_cast<std::chrono::milliseconds>(endTime - conn->startTime);
+    auto time_to_first_byte_ms = std::to_string(duration.count());
+    switch_core_session_t* session = switch_core_session_locate(d->session_id);
+    if (session) {
+      switch_channel_t *channel = switch_core_session_get_channel(session);
+      switch_core_session_rwunlock(session);
+      if (channel) {
+        switch_event_t *event;
+        if (switch_event_create(&event, SWITCH_EVENT_PLAYBACK_START) == SWITCH_STATUS_SUCCESS) {
+          switch_channel_event_set_data(channel, event);
+
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "write_cb: firing playback-started\n");
+
+          switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "Playback-File-Type", "tts_stream");
+          if (d->reported_model_name) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_reported_model_name", d->reported_model_name);
+          }
+          if (d->reported_model_uuid) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_reported_model_uuid", d->reported_model_uuid);
+          }
+          if (d->reported_char_count) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_reported_char_count", d->reported_char_count);
+          }
+          if (d->request_id) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_request_id", d->request_id);
+          }
+          if (d->name_lookup_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_name_lookup_time_ms", d->name_lookup_time_ms);
+          }
+          if (d->connect_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_connect_time_ms", d->connect_time_ms);
+          }
+          if (d->final_response_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_final_response_time_ms", d->final_response_time_ms);
+          }
+          if (d->voice_name) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_voice_name", d->voice_name);
+          }
+          if (d->cache_filename) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", d->cache_filename);
+          }
+
+          switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_time_to_first_byte_ms", time_to_first_byte_ms.c_str());
+          switch_event_fire(&event);
+          d->playback_start_sent = 1;
+        }
+        else {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: failed to create event\n");
+        }
+      }
+      else {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: channel not found\n");
+      }
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: session %s not found\n", d->session_id);
+    }
+  }
+  return size*nmemb;
+}
+
+static bool parseHeader(const std::string& str, std::string& header, std::string& value) {
+    std::vector<std::string> parts;
+    boost::split(parts, str, boost::is_any_of(":"), boost::token_compress_on);
+
+    if (parts.size() != 2)
+        return false;
+
+    header = boost::trim_copy(parts[0]);
+    value = boost::trim_copy(parts[1]);
+    return true;
+}
+
+static int extract_response_code(const std::string& input) {
+  std::size_t space_pos = input.find(' ');
+  if (space_pos == std::string::npos) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Invalid HTTP response format %s\n", input.c_str());
+    return 0;
+  }
+
+  std::size_t code_start_pos = space_pos + 1;
+  std::size_t code_end_pos = input.find(' ', code_start_pos);
+  if (code_end_pos == std::string::npos) {
+    code_end_pos = input.length();
+  }
+
+  std::string code_str = input.substr(code_start_pos, code_end_pos - code_start_pos);
+  int response_code = std::stoi(code_str);
+  return response_code;
+}
+
+static size_t header_callback(char *buffer, size_t size, size_t nitems, ConnInfo_t *conn) {
+  size_t bytes_received = size * nitems;
+  const std::string prefix = "HTTP/";
+  deepgram_t* d = conn->deepgram;
+  std::string header, value;
+  std::string input(buffer, bytes_received);
+  if (parseHeader(input, header, value)) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "recv header: %s with value %s\n", header.c_str(), value.c_str());
+    if (0 == header.compare("dg-request-id")) d->request_id = strdup(value.c_str());
+    else if (0 == header.compare("dg-model-name")) d->reported_model_name = strdup(value.c_str());
+    else if (0 == header.compare("dg-model-uuid")) d->reported_model_uuid = strdup(value.c_str());
+    else if (0 == header.compare("dg-char-count")) d->reported_char_count = strdup(value.c_str());
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: %s\n", input.c_str());
+    if (input.rfind(prefix, 0) == 0) {
+      try {
+        d->response_code = extract_response_code(input);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: parsed response code: %ld\n", d->response_code);
+      } catch (const std::invalid_argument& e) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "header_callback: invalid response code %s\n", input.substr(prefix.length()).c_str());
+      }
+    }
+  }
+  return bytes_received;
+}
+
+/* CURLOPT_OPENSOCKETFUNCTION */
+static curl_socket_t opensocket(void *clientp, curlsocktype purpose, struct curl_sockaddr *address) {
+  curl_socket_t sockfd = CURL_SOCKET_BAD;
+
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "opensocket: %d\n", purpose);
+  /* restrict to IPv4 */
+  if(purpose == CURLSOCKTYPE_IPCXN && address->family == AF_INET) {
+    /* create a tcp socket object */
+    boost::asio::ip::tcp::socket *tcp_socket = new boost::asio::ip::tcp::socket(io_service);
+
+    /* open it and get the native handle*/
+    boost::system::error_code ec;
+    tcp_socket->open(boost::asio::ip::tcp::v4(), ec);
+
+    if(ec) {
+      /* An error occurred */
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't open socket [%ld][%s]\n", ec, ec.message().c_str());
+    }
+    else {
+      sockfd = tcp_socket->native_handle();
+
+      /* save it for monitoring */
+      socket_map.insert(std::pair<curl_socket_t, boost::asio::ip::tcp::socket *>(sockfd, tcp_socket));
+    }
+  }
+  return sockfd;
+}
+
+/* CURLOPT_CLOSESOCKETFUNCTION */
+static int close_socket(void *clientp, curl_socket_t item) {
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "close_socket : %#X\n", item);
+
+  std::map<curl_socket_t, boost::asio::ip::tcp::socket *>::iterator it = socket_map.find(item);
+  if(it != socket_map.end()) {
+    delete it->second;
+    socket_map.erase(it);
+  }
+  return 0;
+}
+
+
+extern "C" {
+  switch_status_t deepgram_speech_load() {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "deepgram_speech_loading..\n");
+    memset(&global, 0, sizeof(GlobalInfo_t));
+    global.multi = curl_multi_init();
+
+     if (!global.multi) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "deepgram_speech_load curl_multi_init() failed, exiting!\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    curl_multi_setopt(global.multi, CURLMOPT_SOCKETFUNCTION, sock_cb);
+    curl_multi_setopt(global.multi, CURLMOPT_SOCKETDATA, &global);
+    curl_multi_setopt(global.multi, CURLMOPT_TIMERFUNCTION, multi_timer_cb);
+    curl_multi_setopt(global.multi, CURLMOPT_TIMERDATA, &global);
+    curl_multi_setopt(global.multi, CURLMOPT_PIPELINING, CURLPIPE_MULTIPLEX);
+
+    /* create temp folder for cache files */
+    const char* baseDir = std::getenv("JAMBONZ_TMP_CACHE_FOLDER");
+    if (!baseDir) {
+      baseDir = "/tmp/";
+    }
+    if (strcmp(baseDir, "/") == 0) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", baseDir);
+      return SWITCH_STATUS_FALSE;
+    }
+
+    fullDirPath = std::string(baseDir) + "tts-cache-files";
+
+    // Create the directory with read, write, and execute permissions for everyone
+    mode_t oldMask = umask(0);
+    int result = mkdir(fullDirPath.c_str(), S_IRWXU | S_IRWXG | S_IRWXO);
+    umask(oldMask);
+    if (result != 0) {
+      if (errno != EEXIST) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", fullDirPath.c_str());
+        fullDirPath = "";
+      }
+      else switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "folder %s already exists\n", fullDirPath.c_str());
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "created folder %s\n", fullDirPath.c_str());
+    }
+
+    /* start worker thread that handles transfers*/
+    std::thread t(threadFunc) ;
+    worker_thread.swap( t ) ;
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "deepgram_speech_loaded..\n");
+
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t deepgram_speech_unload() {
+    /* stop the ASIO IO service */
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "deepgram_speech_unload: stopping io service\n");
+    io_service.stop();
+
+    /* Join the worker thread */
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "deepgram_speech_unload: wait for worker thread to complete\n");
+    if (worker_thread.joinable()) {
+        worker_thread.join();
+    }
+
+    /* cleanup curl multi handle*/
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "deepgram_speech_unload: release curl multi\n");
+    curl_multi_cleanup(global.multi);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "deepgram_speech_unload: completed\n");
+
+		return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t deepgram_speech_open(deepgram_t* deepgram) {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t deepgram_speech_feed_tts(deepgram_t* d, char* text, switch_speech_flag_t *flags) {
+    CURLMcode rc;
+
+    const int MAX_CHARS = 20;
+    char tempText[MAX_CHARS + 4]; // +4 for the ellipsis and null terminator
+
+    if (strlen(text) > MAX_CHARS) {
+        strncpy(tempText, text, MAX_CHARS);
+        strcpy(tempText + MAX_CHARS, "...");
+    } else {
+        strcpy(tempText, text);
+    }
+
+    /* open cache file */
+    if (d->cache_audio && fullDirPath.length() > 0) {
+      switch_uuid_t uuid;
+      char uuid_str[SWITCH_UUID_FORMATTED_LENGTH + 1];
+      char outfile[512] = "";
+      int fd;
+
+      switch_uuid_get(&uuid);
+      switch_uuid_format(uuid_str, &uuid);
+
+      switch_snprintf(outfile, sizeof(outfile), "%s%s%s.r8", fullDirPath.c_str(), SWITCH_PATH_SEPARATOR, uuid_str);
+      d->cache_filename = strdup(outfile);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "writing audio cache file to %s\n", d->cache_filename);
+
+      mode_t oldMask = umask(0);
+      fd = open(outfile, O_WRONLY | O_CREAT | O_TRUNC, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH);
+      umask(oldMask);
+      if (fd == -1 ) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+      }
+      else {
+        d->file = fdopen(fd, "wb");
+        if (!d->file) {
+          close(fd);
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+        }
+      }
+    }
+
+    if (!d->api_key) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "deepgram_speech_feed_tts: no api_key provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+    /* format url*/
+    std::string url;
+    std::ostringstream url_stream;
+    // always use sample_rate=8000 for support jambonz caching system.
+    url_stream << (d->endpoint != nullptr ? d->endpoint : "https://api.deepgram.com") << "/v1/speak?model=" << d->voice_name << "&encoding=linear16&sample_rate=8000";
+    url = url_stream.str();
+
+    /* create the JSON body */
+    cJSON * jResult = cJSON_CreateObject();
+    cJSON_AddStringToObject(jResult, "text", text);
+    char *json = cJSON_PrintUnformatted(jResult);
+
+    cJSON_Delete(jResult);
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "deepgram_speech_feed_tts: [%s] [%s]\n", url.c_str(), tempText);
+
+    ConnInfo_t *conn = pool.malloc() ;
+
+    CURL* easy = createEasyHandle();
+    d->conn = (void *) conn ;
+    conn->deepgram = d;
+    conn->easy = easy;
+    conn->global = &global;
+    conn->hdr_list = NULL ;
+    conn->file = d->file;
+    conn->body = json;
+    conn->flushed = false;
+    conn->has_last_byte = false;
+    conn->last_byte = 0;
+
+
+    d->circularBuffer = (void *) new CircularBuffer_t(BUFFER_GROW_SIZE);
+    // Always use deepgram at rate 8000 for helping cache audio from jambonz.
+    if (d->rate != 8000) {
+      int err;
+      d->resampler = speex_resampler_init(1, 8000, d->rate, SWITCH_RESAMPLE_QUALITY, &err);
+      if (0 != err) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing resampler: %s.\n", speex_resampler_strerror(err));
+        return SWITCH_STATUS_FALSE;
+      }
+    }
+
+    std::ostringstream api_key_stream;
+    api_key_stream << "Authorization: Token " << d->api_key;
+
+    curl_easy_setopt(easy, CURLOPT_URL, url.c_str());
+    curl_easy_setopt(easy, CURLOPT_WRITEFUNCTION, write_cb);
+    curl_easy_setopt(easy, CURLOPT_WRITEDATA, conn);
+    curl_easy_setopt(easy, CURLOPT_ERRORBUFFER, conn->error);
+    curl_easy_setopt(easy, CURLOPT_PRIVATE, conn);
+    curl_easy_setopt(easy, CURLOPT_VERBOSE, 0L);
+    curl_easy_setopt(easy, CURLOPT_NOPROGRESS, 1L);
+    curl_easy_setopt(easy, CURLOPT_HEADERFUNCTION, header_callback);
+    curl_easy_setopt(easy, CURLOPT_HEADERDATA, conn);
+    
+    /* call this function to get a socket */
+    curl_easy_setopt(easy, CURLOPT_OPENSOCKETFUNCTION, opensocket);
+
+    /* call this function to close a socket */
+    curl_easy_setopt(easy, CURLOPT_CLOSESOCKETFUNCTION, close_socket);
+
+    conn->hdr_list = curl_slist_append(conn->hdr_list, api_key_stream.str().c_str());
+    conn->hdr_list = curl_slist_append(conn->hdr_list, "Content-Type: application/json");
+    curl_easy_setopt(easy, CURLOPT_HTTPHEADER, conn->hdr_list);
+
+    curl_easy_setopt(easy, CURLOPT_POSTFIELDS, conn->body);
+    //curl_easy_setopt(easy, CURLOPT_POSTFIELDSIZE, body.length());
+
+    // libcurl adding random byte to the response body that creates white noise to audio file
+    // https://github.com/curl/curl/issues/10525
+    const bool disable_http_2 = switch_true(std::getenv("DISABLE_HTTP2_FOR_TTS_STREAMING"));
+    curl_easy_setopt(easy, CURLOPT_HTTP_VERSION, disable_http_2 ? CURL_HTTP_VERSION_1_1 : CURL_HTTP_VERSION_2_0);
+
+    rc = curl_multi_add_handle(global.multi, conn->easy);
+    mcode_test("new_conn: curl_multi_add_handle", rc);
+
+    /* start a timer to measure the duration until we receive first byte of audio */
+    conn->startTime = std::chrono::high_resolution_clock::now();
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "deepgram_speech_feed_tts: called curl_multi_add_handle\n");
+
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t deepgram_speech_read_tts(deepgram_t* d, void *data, size_t *datalen, switch_speech_flag_t *flags) {
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) d->circularBuffer;
+    std::vector<uint16_t> pcm_data;
+
+    {
+      switch_mutex_lock(d->mutex);
+      ConnInfo_t *conn = (ConnInfo_t *) d->conn;
+      if (d->response_code > 0 && d->response_code != 200) {
+        switch_mutex_unlock(d->mutex);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "deepgram_speech_read_tts, returning failure\n") ;  
+        return SWITCH_STATUS_FALSE;
+      }
+      if (conn && conn->flushed) {
+        switch_mutex_unlock(d->mutex);
+        return SWITCH_STATUS_BREAK;
+      }
+      if (cBuffer->empty()) {
+        if (d->draining) {
+          switch_mutex_unlock(d->mutex);
+          return SWITCH_STATUS_BREAK;
+        }
+        /* no audio available yet so send silence */
+        memset(data, 255, *datalen);
+        switch_mutex_unlock(d->mutex);
+        return SWITCH_STATUS_SUCCESS;
+      }
+      size_t size = std::min((*datalen/(2 * d->rate / 8000)), cBuffer->size());
+      pcm_data.insert(pcm_data.end(), cBuffer->begin(), cBuffer->begin() + size);
+      cBuffer->erase(cBuffer->begin(), cBuffer->begin() + size);
+      switch_mutex_unlock(d->mutex);
+    }
+
+    size_t data_size = pcm_data.size();
+
+    if (d->resampler) {
+      std::vector<int16_t> in(pcm_data.begin(), pcm_data.end());
+
+      std::vector<int16_t> out((*datalen));
+      spx_uint32_t in_len = data_size;
+      spx_uint32_t out_len = out.size();
+      speex_resampler_process_interleaved_int(d->resampler, in.data(), &in_len, out.data(), &out_len);
+
+      if (out_len > out.size()) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, "Resampler output exceeded maximum buffer size!\n");
+        return SWITCH_STATUS_FALSE;
+      }
+
+      memcpy(data, out.data(), out_len * sizeof(int16_t));
+      *datalen = out_len * sizeof(int16_t);
+    } else {
+      memcpy(data, pcm_data.data(), pcm_data.size() * sizeof(uint16_t));
+      *datalen = pcm_data.size() * sizeof(uint16_t);
+    }
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t deepgram_speech_flush_tts(deepgram_t* d) {
+    bool download_complete = d->response_code == 200;
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "deepgram_speech_flush_tts, download complete? %s\n", download_complete ? "yes" : "no") ;  
+
+    ConnInfo_t *conn = (ConnInfo_t *) d->conn;
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) d->circularBuffer;
+    delete cBuffer;
+    d->circularBuffer = nullptr ;
+
+    // destroy resampler
+    if (d->resampler) {
+      speex_resampler_destroy(d->resampler);
+      d->resampler = NULL;
+    }
+
+    if (conn) {
+      conn->flushed = true;
+      if (!download_complete) {
+        if (conn->file) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "closing audio cache file %s because download was interrupted\n", d->cache_filename);
+          if (fclose(conn->file) != 0) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "error closing audio cache file\n");
+          }
+          conn->file = nullptr ;
+        }
+      }
+    }
+    // If playback_start has not been sent, delete the file
+    if (d->cache_filename && !d->playback_start_sent) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing audio cache file %s because download was interrupted\n", d->cache_filename);
+      if (unlink(d->cache_filename) != 0) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error removing audio cache file %s: %d:%s\n", 
+          d->cache_filename, errno, strerror(errno));
+      }
+      free(d->cache_filename);
+      d->cache_filename = nullptr ;
+    }
+    if (d->session_id) {
+      switch_core_session_t* session = switch_core_session_locate(d->session_id);
+      if (session) {
+        switch_channel_t *channel = switch_core_session_get_channel(session);
+        switch_core_session_rwunlock(session);
+        if (channel) {
+          switch_event_t *event;
+          if (switch_event_create(&event, SWITCH_EVENT_PLAYBACK_STOP) == SWITCH_STATUS_SUCCESS) {
+            switch_channel_event_set_data(channel, event);
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "Playback-File-Type", "tts_stream");
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_deepgram_response_code", std::to_string(d->response_code).c_str());
+            if (d->cache_filename && d->response_code == 200) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", d->cache_filename);
+            }
+            if (d->response_code != 200 && d->err_msg) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_error", d->err_msg);
+            }
+            switch_event_fire(&event);
+          }
+          else {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: failed to create event\n");
+          }
+        }
+        else {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: channel not found\n");
+        }
+      }
+    }
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+	switch_status_t deepgram_speech_close(deepgram_t* w) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "deepgram_speech_close\n") ;
+		return SWITCH_STATUS_SUCCESS;
+	}
+}

mod_deepgram_tts/deepgram_glue.h

@@ -0,0 +1,12 @@
+#ifndef __DEEPGRAM_GLUE_H__
+#define __DEEPGRAM_GLUE_H__
+
+switch_status_t deepgram_speech_load();
+switch_status_t deepgram_speech_open(deepgram_t* deepgram);
+switch_status_t deepgram_speech_feed_tts(deepgram_t* deepgram, char* text, switch_speech_flag_t *flags);
+switch_status_t deepgram_speech_read_tts(deepgram_t* deepgram, void *data, size_t *datalen, switch_speech_flag_t *flags);
+switch_status_t deepgram_speech_flush_tts(deepgram_t* deepgram);
+switch_status_t deepgram_speech_close(deepgram_t* deepgram);
+switch_status_t deepgram_speech_unload();
+
+#endif

mod_deepgram_tts/mod_deepgram_tts.c

@@ -0,0 +1,168 @@
+#include "mod_deepgram_tts.h"
+#include "deepgram_glue.h"
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_deepgram_tts_load);
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_deepgram_tts_shutdown);
+SWITCH_MODULE_DEFINITION(mod_deepgram_tts, mod_deepgram_tts_load, mod_deepgram_tts_shutdown, NULL);
+
+static void cleardeepgram(deepgram_t* d, int freeAll) {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "cleardeepgram\n");
+  if (d->api_key) free(d->api_key);
+
+  if (d->request_id) free(d->request_id);
+  if (d->reported_model_name) free(d->reported_model_name);
+  if (d->reported_model_uuid) free(d->reported_model_uuid);
+  if (d->reported_char_count) free(d->reported_char_count);
+  if (d->ct) free(d->ct);
+  if (d->err_msg) free(d->err_msg);
+  if (d->name_lookup_time_ms) free(d->name_lookup_time_ms);
+  if (d->connect_time_ms) free(d->connect_time_ms);
+  if (d->final_response_time_ms) free(d->final_response_time_ms);
+  if (d->cache_filename) free(d->cache_filename);
+  if (d->endpoint) free(d->endpoint);
+  
+
+  d->api_key = NULL;
+  d->request_id = NULL;
+  d->endpoint = NULL;
+
+  d->reported_model_name = NULL;
+  d->reported_model_uuid = NULL;
+  d->reported_char_count = NULL;
+  d->ct = NULL;
+  d->err_msg = NULL;
+  d->name_lookup_time_ms = NULL;
+  d->connect_time_ms = NULL;
+  d->final_response_time_ms = NULL;
+  d->cache_filename = NULL;
+
+  if (freeAll) {
+    if (d->voice_name) free(d->voice_name);
+    if (d->session_id) free(d->session_id);
+    d->voice_name = NULL;
+    d->session_id = NULL;
+  }
+}
+
+static deepgram_t * createOrRetrievePrivateData(switch_speech_handle_t *sh) {
+  deepgram_t *d = (deepgram_t *) sh->private_info;  
+  if (!d) {
+    d = switch_core_alloc(sh->memory_pool, sizeof(*d));
+  	sh->private_info = d;
+    memset(d, 0, sizeof(*d));
+    switch_mutex_init(&d->mutex, SWITCH_MUTEX_NESTED, sh->memory_pool);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "allocated deepgram_t\n");
+  }
+  return d;
+}
+
+switch_status_t d_speech_open(switch_speech_handle_t *sh, const char *voice_name, int rate, int channels, switch_speech_flag_t *flags)
+{
+  deepgram_t *d = createOrRetrievePrivateData(sh);
+  d->voice_name = strdup(voice_name);
+  d->rate = rate;
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "d_speech_open voice: %s, rate %d, channels %d\n", voice_name, rate, channels);
+  return deepgram_speech_open(d);
+}
+
+static switch_status_t d_speech_close(switch_speech_handle_t *sh, switch_speech_flag_t *flags)
+{
+  switch_status_t rc;
+  deepgram_t *d = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "d_speech_close\n");
+
+  switch_mutex_destroy(d->mutex);
+
+  rc = deepgram_speech_close(d);
+  cleardeepgram(d, 1);
+  return rc;
+}
+
+/**
+ * Freeswitch will call this function to feed us text to speak
+ */
+static switch_status_t d_speech_feed_tts(switch_speech_handle_t *sh, char *text, switch_speech_flag_t *flags)
+{
+  deepgram_t *d = createOrRetrievePrivateData(sh);
+  d->draining = 0;
+  d->reads = 0;
+  d->response_code = 0;
+  d->err_msg = NULL;
+  d->playback_start_sent = 0;
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "d_speech_feed_tts\n");
+
+  return deepgram_speech_feed_tts(d, text, flags);
+}
+
+/**
+ * Freeswitch calls periodically to get some rendered audio in L16 format. We can provide up to 8k of audio at a time.
+ */
+static switch_status_t d_speech_read_tts(switch_speech_handle_t *sh, void *data, size_t *datalen, switch_speech_flag_t *flags)
+{
+  deepgram_t *d = createOrRetrievePrivateData(sh);
+  return deepgram_speech_read_tts(d, data, datalen, flags);
+}
+
+/**
+ * This is called at the end, not sure exactly what we need to do here..
+ */
+static void d_speech_flush_tts(switch_speech_handle_t *sh)
+{
+  deepgram_t *d = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "d_speech_flush_tts\n");
+  deepgram_speech_flush_tts(d);
+
+  cleardeepgram(d, 0);
+}
+
+static void d_text_param_tts(switch_speech_handle_t *sh, char *param, const char *val)
+{
+  deepgram_t *d = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "d_text_param_tts: %s=%s\n", param, val);
+  if (0 == strcmp(param, "api_key")) {
+    if (d->api_key) free(d->api_key);
+    d->api_key = strdup(val);
+  } else if (0 == strcmp(param, "endpoint")) {
+    if (d->endpoint) free(d->endpoint);
+    d->endpoint = strdup(val);
+  } else if (0 == strcmp(param, "voice")) {
+    if (d->voice_name) free(d->voice_name);
+    d->voice_name = strdup(val);
+  } else if (0 == strcmp(param, "session-uuid")) {
+    if (d->session_id) free(d->session_id);
+    d->session_id = strdup(val);
+  } else if (0 == strcmp(param, "write_cache_file") && switch_true(val)) {
+    d->cache_audio = 1;
+  }
+}
+
+static void d_numeric_param_tts(switch_speech_handle_t *sh, char *param, int val)
+{
+}
+static void d_float_param_tts(switch_speech_handle_t *sh, char *param, double val)
+{
+}
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_deepgram_tts_load)
+{
+  switch_speech_interface_t *speech_interface;
+
+  *module_interface = switch_loadable_module_create_module_interface(pool, modname);
+  speech_interface = switch_loadable_module_create_interface(*module_interface, SWITCH_SPEECH_INTERFACE);
+  speech_interface->interface_name = "deepgram";
+  speech_interface->speech_open = d_speech_open;
+  speech_interface->speech_close = d_speech_close;
+  speech_interface->speech_feed_tts = d_speech_feed_tts;
+  speech_interface->speech_read_tts = d_speech_read_tts;
+	speech_interface->speech_flush_tts = d_speech_flush_tts;
+	speech_interface->speech_text_param_tts = d_text_param_tts;
+	speech_interface->speech_numeric_param_tts = d_numeric_param_tts;
+	speech_interface->speech_float_param_tts = d_float_param_tts;
+  return deepgram_speech_load();
+}
+
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_deepgram_tts_shutdown)
+{
+  return deepgram_speech_unload();
+}

mod_deepgram_tts/mod_deepgram_tts.h

@@ -0,0 +1,44 @@
+#ifndef __MOD_DEEPGRAM_TTS_H__
+#define __MOD_DEEPGRAM_TTS_H__
+
+#include <switch.h>
+#include <speex/speex_resampler.h>
+
+typedef struct deepgram_data {
+  char *voice_name;
+  char *api_key;
+  char *endpoint;
+
+  /* result data */
+  long response_code;
+  char *ct;
+  // Deepgram hedaers
+  //dg-model-name
+  char *reported_model_name;
+  //dg-model-uuid
+  char *reported_model_uuid;
+  //dg-char-count
+  char *reported_char_count;
+  //dg-request-id
+  char *request_id;
+  char *name_lookup_time_ms;
+  char *connect_time_ms;
+  char *final_response_time_ms;
+  char *err_msg;
+  char *cache_filename;
+  char *session_id;
+
+  int rate;
+  int draining;
+  int reads;
+  int cache_audio;
+  int playback_start_sent;
+
+	void *conn;
+  void *circularBuffer;
+  switch_mutex_t *mutex;
+  FILE *file;
+  SpeexResamplerState *resampler;
+} deepgram_t;
+
+#endif

mod_dialogflow/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_dub/Makefile.am

@@ -0,0 +1,18 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_dub
+
+mod_LTLIBRARIES = mod_dub.la
+mod_dub_la_SOURCES  = ap_file.cpp ap_http.cpp track.cpp dub_glue.cpp tts_vendor_parser.cpp mod_dub.c vector_math.cpp mpg_decode.cpp
+mod_dub_la_CFLAGS   = $(AM_CFLAGS)
+mod_dub_la_CXXFLAGS = $(AM_CXXFLAGS) -std=c++17
+
+if USE_AVX2
+mod_dub_la_CXXFLAGS += -mavx2 -DUSE_AVX2
+else
+if USE_SSE2
+mod_dub_la_CXXFLAGS += -msse2 -DUSE_SSE2
+endif
+endif
+
+mod_dub_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_dub_la_LDFLAGS  = -avoid-version -module -no-undefined -lstdc++ -lmpg123 -lboost_system -lboost_thread

mod_dub/ap.h

@@ -0,0 +1,41 @@
+#ifndef __AP_H__
+#define __AP_H__
+
+#include <mutex>
+#include <functional>
+#include "common.h"
+
+class AudioProducer {
+public:
+  AudioProducer(
+    std::mutex& mutex,
+    CircularBuffer_t& circularBuffer,
+    int sampleRate
+  ) : _mutex(mutex), _buffer(circularBuffer), _sampleRate(sampleRate), _notified(false), _loop(false), _gain(0) {}
+  virtual ~AudioProducer() {}
+
+  virtual void notifyDone(bool error, const std::string& errorMsg) {
+    if (!_notified) {
+      _notified = true;
+      if (_callback) _callback(error, errorMsg);
+    }
+  }
+  virtual void start(std::function<void(bool, const std::string&)> callback) = 0;
+  virtual void stop() = 0;
+
+  bool isLoopedAudio() const { return _loop; }
+
+protected:
+  std::mutex& _mutex;
+  CircularBuffer_t& _buffer;
+  int _sampleRate;
+  int _gain;
+  bool _loop;
+  std::function<void(bool, const std::string&)> _callback;
+  bool _notified;
+};
+
+
+
+
+#endif

mod_dub/ap_file.cpp

@@ -0,0 +1,199 @@
+#include "switch.h"
+#include "ap_file.h"
+#include "mpg_decode.h"
+
+#define INIT_BUFFER_SIZE (80000)
+#define BUFFER_GROW_SIZE (80000)
+#define BUFFER_THROTTLE_LOW (40000)
+#define BUFFER_THROTTLE_HIGH (160000)
+
+bool AudioProducerFile::initialized = false;
+boost::asio::io_service AudioProducerFile::io_service;
+std::thread AudioProducerFile::worker_thread;
+
+void AudioProducerFile::threadFunc() {      
+  io_service.reset() ;
+  boost::asio::io_service::work work(io_service);
+  
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "file_loader threadFunc - starting\n");
+
+  for(;;) {
+      
+    try {
+      io_service.run() ;
+      break ;
+    }
+    catch( std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "file_loader threadFunc - Error: %s\n", e.what());
+    }
+  }
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "file_loader threadFunc - ending\n");
+}
+
+void AudioProducerFile::_init() {
+  if (!initialized) {
+    initialized = true;
+    if (mpg123_init() != MPG123_OK) {
+      throw std::runtime_error("AudioProducerFile::AudioProducerFile: failed to initiate MPG123");
+      return ;
+    }
+
+    /* start worker thread */
+    std::thread t(threadFunc) ;
+    worker_thread.swap( t ) ;
+  }
+}
+
+void AudioProducerFile::_deinit() {
+  if (initialized) {
+    initialized = false;
+    io_service.stop();
+    if (worker_thread.joinable()) {
+      worker_thread.join();
+    }
+    mpg123_exit();
+  }
+}
+AudioProducerFile::AudioProducerFile(
+  std::mutex& mutex,
+  CircularBuffer_t& circularBuffer,
+  int sampleRate
+) : AudioProducer(mutex, circularBuffer, sampleRate), _timer(io_service), _mh(nullptr), _fp(nullptr) {
+
+  AudioProducerFile::_init();
+}
+
+AudioProducerFile::~AudioProducerFile() {
+  reset();
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "AudioProducerFile::~AudioProducerFile %p\n", (void *)this);
+}
+
+void AudioProducerFile::queueFileAudio(const std::string& path, int gain, bool loop) {
+    _path = path;
+    _gain = gain;
+    _loop = loop;
+
+    /* we only handle mp3 or r8 files atm */
+    size_t pos = path.find_last_of('.');
+    if (pos == std::string::npos) {
+      throw std::runtime_error("file " + path + " has no extension");
+    }
+    auto filetype = path.substr(pos + 1);
+    if (0 == filetype.compare("mp3")) _type = FILE_TYPE_MP3;
+    else if (0 == filetype.compare("r8")) _type = FILE_TYPE_R8;
+    else throw std::runtime_error("file " + path + " has unsupported extension " + filetype);
+}
+
+void AudioProducerFile::start(std::function<void(bool, const std::string&)> callback) {
+    int mhError = 0;
+
+    _callback = callback;
+
+    /* allocate handle for mpeg decoding */
+    _mh = mpg123_new("auto", &mhError);
+    if (!_mh) {
+      const char *mhErr = mpg123_plain_strerror(mhError);
+      throw std::runtime_error("Error allocating mpg123 handle! " + std::string(mhErr));
+    }
+
+    if (mpg123_open_feed(_mh) != MPG123_OK) throw std::runtime_error("Error mpg123_open_feed!");
+    if (mpg123_format_all(_mh) != MPG123_OK) throw std::runtime_error("Error mpg123_format_all!");
+    if (mpg123_param(_mh, MPG123_FORCE_RATE, _sampleRate, 0) != MPG123_OK) throw std::runtime_error("Error forcing resample to 8k!");
+    if (mpg123_param(_mh, MPG123_FLAGS, MPG123_MONO_MIX, 0) != MPG123_OK) throw std::runtime_error("Error forcing single channel!");
+    
+    _fp = fopen(_path.c_str(), "rb");
+    if (!_fp) throw std::runtime_error("Error opening file " + _path);
+
+    _status = Status_t::STATUS_AWAITING_RESTART;
+
+    /* do the initial read in the worker thread so we don't block here */
+    _timer.expires_from_now(boost::posix_time::millisec(1));
+    _timer.async_wait(boost::bind(&AudioProducerFile::read_cb, this, boost::placeholders::_1));
+}
+
+void AudioProducerFile::read_cb(const boost::system::error_code& error) {
+  if (_status == Status_t::STATUS_STOPPED) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "read_cb: session gone\n");
+    return;
+  }
+  if (_status == Status_t::STATUS_AWAITING_RESTART) {
+    _status = Status_t::STATUS_IN_PROGRESS;
+  }
+  if (!error) {
+    size_t size = _buffer.size();
+    if (size < BUFFER_THROTTLE_LOW) {
+      std::vector<int16_t> pcm_data;
+      int8_t buf[INIT_BUFFER_SIZE];
+
+      size_t bytesRead = ::fread(buf, sizeof(int8_t), INIT_BUFFER_SIZE, _fp);
+      if (bytesRead <= 0) {
+        if (::feof(_fp)) switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "read_cb: %p eof\n", (void *) this);
+        else if (::ferror(_fp)) switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "read_cb: %p error reading file\n", (void *) this);
+        else switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "read_cb: %p unknown error reading file\n", (void *) this);
+        _status = Status_t::STATUS_COMPLETE;
+        return;
+      }
+
+      if (_type == FileType_t::FILE_TYPE_MP3) pcm_data = convert_mp3_to_linear(_mh, _gain, buf, bytesRead);
+      else pcm_data = std::vector<int16_t>(reinterpret_cast<int16_t*>(buf), reinterpret_cast<int16_t*>(buf) + bytesRead / 2);
+
+      {
+        std::lock_guard<std::mutex> lock(_mutex); 
+
+        // Resize the buffer if necessary
+        size_t capacity = _buffer.capacity();
+        if (capacity - size < pcm_data.size()) {
+          //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "write_cb %p growing buffer, size now %ld\n", (void *) this, size); 
+          _buffer.set_capacity(size + std::max(pcm_data.size(), (size_t)BUFFER_GROW_SIZE));
+        }
+        
+        /* Push the data into the buffer */
+        _buffer.insert(_buffer.end(), pcm_data.data(), pcm_data.data() + pcm_data.size());
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "read_cb: %p wrote data, buffer size is now %ld\n", (void *) this, _buffer.size());        
+      }
+
+      if (bytesRead < INIT_BUFFER_SIZE) {
+        _status = Status_t::STATUS_COMPLETE;
+        //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "read_cb: %u reached end of file, status is %s\n", (void *) this, status2String(_status));
+      }
+    }
+
+    if (_status == Status_t::STATUS_COMPLETE) {
+      cleanup(Status_t::STATUS_COMPLETE);
+    }
+    else {
+      // read more in 2 seconds
+      _timer.expires_from_now(boost::posix_time::millisec(2000));
+      _timer.async_wait(boost::bind(&AudioProducerFile::read_cb, this, boost::placeholders::_1));
+    }
+  } else {
+    cleanup(Status_t::STATUS_FAILED, error.message());
+  }
+}
+
+void AudioProducerFile::stop() {
+  cleanup(Status_t::STATUS_STOPPED, "");
+}
+
+void AudioProducerFile::reset() {
+  {
+    std::lock_guard<std::mutex> lock(_mutex);
+    if (_fp) {
+      fclose(_fp);
+      _fp = nullptr;
+    }
+    if (_mh) {
+      mpg123_close(_mh);
+      mpg123_delete(_mh);
+      _mh = nullptr;
+    }
+  }
+  _timer.cancel();
+  _status = Status_t::STATUS_NONE;
+}
+
+void AudioProducerFile::cleanup(Status_t status, std::string errMsg) {
+  reset();
+  _status = status;
+  notifyDone(status != Status_t::STATUS_COMPLETE, errMsg);
+}

mod_dub/ap_file.h

@@ -0,0 +1,89 @@
+#ifndef __AP_FILE_H__
+#define __AP_FILE_H__
+
+#include <thread>
+#include <boost/asio.hpp>
+#include <boost/bind/bind.hpp>
+
+#include <mpg123.h>
+
+#include "ap.h"
+
+class AudioProducerFile : public AudioProducer {
+public:
+
+  typedef enum
+  {
+    STATUS_NONE = 0,
+    STATUS_FAILED,
+    STATUS_IN_PROGRESS,
+    STATUS_PAUSED,
+    STATUS_COMPLETE,
+    STATUS_AWAITING_RESTART,
+    STATUS_STOPPED
+  } Status_t;
+
+  typedef enum {
+    FILE_TYPE_MP3 = 0,
+    FILE_TYPE_R8
+  } FileType_t;
+
+  const char* status2String(Status_t status)
+  {
+    static const char* statusStrings[] = {
+      "STATUS_NONE",
+      "STATUS_FAILED",
+      "STATUS_IN_PROGRESS",
+      "STATUS_PAUSED",
+      "STATUS_COMPLETE",
+      "STATUS_AWAITING_RESTART",
+      "STATUS_STOPPED"
+    };
+
+    if (status >= 0 && status < sizeof(statusStrings) / sizeof(statusStrings[0]))
+    {
+      return statusStrings[status];
+    }
+    else
+    {
+      return "UNKNOWN_STATUS";
+    }
+  }
+
+  AudioProducerFile(
+    std::mutex& mutex,
+    CircularBuffer_t& circularBuffer,
+    int sampleRate
+  );
+  virtual ~AudioProducerFile();
+  virtual void start(std::function<void(bool, const std::string&)> callback);
+  virtual void stop();
+  void cleanup(Status_t status, std::string errMsg = "");
+  void reset();
+
+  void queueFileAudio(const std::string& path, int gain = 0, bool loop = false);
+
+  void read_cb(const boost::system::error_code& error);
+
+  static bool initialized;
+  static std::thread worker_thread;
+  static boost::asio::io_service io_service;
+  static void threadFunc();
+
+private:
+
+  static void _init();
+  static void _deinit();
+
+  void stop_file_load();
+
+  std::string _path;
+  Status_t _status;
+  FileType_t _type;
+  mpg123_handle *_mh;
+  boost::asio::deadline_timer _timer;
+  FILE* _fp;
+
+};
+
+#endif

mod_dub/ap_http.cpp

@@ -0,0 +1,669 @@
+#include <functional>
+#include <mpg123.h>
+#include "switch.h"
+#include "ap_http.h"
+#include "mpg_decode.h"
+
+#include <boost/asio/ssl.hpp>
+#include <boost/pool/object_pool.hpp>
+#include <boost/bind/bind.hpp>
+#include <boost/tokenizer.hpp>
+#include <boost/algorithm/string.hpp>
+#include <boost/assign/list_of.hpp>
+
+#define BUFFER_GROW_SIZE (80000)
+#define BUFFER_THROTTLE_LOW (40000)
+#define BUFFER_THROTTLE_HIGH (160000)
+
+bool AudioProducerHttp::initialized = false;
+boost::asio::io_service AudioProducerHttp::io_service;
+std::thread AudioProducerHttp::worker_thread;
+GlobalInfo_t AudioProducerHttp::global;
+std::map<curl_socket_t, boost::asio::ip::tcp::socket *> AudioProducerHttp::socket_map;
+boost::asio::deadline_timer AudioProducerHttp::multi_timer(AudioProducerHttp::io_service);
+
+void AudioProducerHttp::threadFunc() {      
+  /* to make sure the event loop doesn't terminate when there is no work to do */
+  io_service.reset() ;
+  boost::asio::io_service::work work(io_service);
+  
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "ap_http threadFunc - starting\n");
+
+  for(;;) {
+      
+    try {
+      io_service.run() ;
+      break ;
+    }
+    catch( std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "ap_http threadFunc - Error: %s\n", e.what());
+    }
+  }
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "ap_http threadFunc - ending\n");
+}
+
+void AudioProducerHttp::_init() {
+  if (!initialized) {
+    initialized = true;
+    if (mpg123_init() != MPG123_OK) {
+      throw std::runtime_error("AudioProducerFile::AudioProducerFile: failed to initiate MPG123");
+      return ;
+    }
+
+    memset(&global, 0, sizeof(GlobalInfo_t));
+    global.multi = curl_multi_init();
+    if (!global.multi) {
+      throw std::runtime_error("AudioProducerHttp::_init: failed to initiate CURL multi");
+      return ;
+    }
+
+    curl_multi_setopt(global.multi, CURLMOPT_SOCKETFUNCTION, &AudioProducerHttp::sock_cb);
+    curl_multi_setopt(global.multi, CURLMOPT_SOCKETDATA, &global);
+    curl_multi_setopt(global.multi, CURLMOPT_TIMERFUNCTION, &AudioProducerHttp::multi_timer_cb);
+    curl_multi_setopt(global.multi, CURLMOPT_TIMERDATA, &global);
+    curl_multi_setopt(global.multi, CURLMOPT_PIPELINING, CURLPIPE_MULTIPLEX);
+
+    /* start worker thread */
+    std::thread t(threadFunc) ;
+    worker_thread.swap( t ) ;
+  }
+}
+
+void AudioProducerHttp::_deinit() {
+  if (initialized) {
+    initialized = false;
+    io_service.stop();
+    if (worker_thread.joinable()) {
+      worker_thread.join();
+    }
+    /* cleanup curl multi handle*/
+    curl_multi_cleanup(global.multi);
+
+    mpg123_exit();
+  }
+}
+
+AudioProducerHttp::AudioProducerHttp(
+    std::mutex& mutex,
+    CircularBuffer_t& circularBuffer,
+    int sampleRate
+) : AudioProducer(mutex, circularBuffer, sampleRate), _status(Status_t::STATUS_NONE), _mh(nullptr), _easy(nullptr), 
+    _error{0}, _response_code(0), _timer(io_service) {
+
+  AudioProducerHttp::_init();
+}
+
+AudioProducerHttp::~AudioProducerHttp() {
+  reset();
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "AudioProducerFile::~AudioProducerFile %p\n", (void *)this);
+}
+
+void AudioProducerHttp::start(std::function<void(bool, const std::string&)> callback) {
+  int mhError = 0;
+
+  _callback = callback;
+  memset(_error, 0, sizeof(_error));
+
+  /* allocate handle for mpeg decoding */
+  _mh = mpg123_new("auto", &mhError);
+  if (!_mh) {
+    const char *mhErr = mpg123_plain_strerror(mhError);
+    throw std::runtime_error("Error allocating mpg123 handle! " + std::string(mhErr));
+  }
+
+  if (mpg123_open_feed(_mh) != MPG123_OK) throw std::runtime_error("Error mpg123_open_feed!\n");
+  if (mpg123_format_all(_mh) != MPG123_OK) throw std::runtime_error("Error mpg123_format_all!\n");
+  if (mpg123_param(_mh, MPG123_FORCE_RATE, _sampleRate, 0) != MPG123_OK) throw std::runtime_error("Error forcing resample to 8k!\n");
+  if (mpg123_param(_mh, MPG123_FLAGS, MPG123_MONO_MIX, 0) != MPG123_OK) throw std::runtime_error("Error forcing single channel!\n");
+
+  _easy = createEasyHandle();
+  if (!_easy) throw std::runtime_error("Error creating easy handle!\n");
+
+  curl_easy_setopt(_easy, CURLOPT_WRITEFUNCTION, &AudioProducerHttp::static_write_cb);
+  curl_easy_setopt(_easy, CURLOPT_URL, _url.c_str());
+  curl_easy_setopt(_easy, CURLOPT_HTTPGET, _method == HttpMethod_t::HTTP_METHOD_GET ? 1L : 0L);
+  curl_easy_setopt(_easy, CURLOPT_WRITEDATA, this);
+  curl_easy_setopt(_easy, CURLOPT_ERRORBUFFER, _error);
+  curl_easy_setopt(_easy, CURLOPT_PRIVATE, this);
+  curl_easy_setopt(_easy, CURLOPT_VERBOSE, 0L);
+  curl_easy_setopt(_easy, CURLOPT_NOPROGRESS, 1L);
+  curl_easy_setopt(_easy, CURLOPT_HEADERFUNCTION, &AudioProducerHttp::static_header_callback);
+  curl_easy_setopt(_easy, CURLOPT_HEADERDATA, this);
+  
+  /* call this function to get a socket */
+  curl_easy_setopt(_easy, CURLOPT_OPENSOCKETFUNCTION, &AudioProducerHttp::open_socket);
+
+  /* call this function to close a socket */
+  curl_easy_setopt(_easy, CURLOPT_CLOSESOCKETFUNCTION, &AudioProducerHttp::close_socket);
+
+  // libcurl adding random byte to the response body that creates white noise to audio file
+  // https://github.com/curl/curl/issues/10525
+  const bool disable_http_2 = switch_true(std::getenv("DISABLE_HTTP2_FOR_TTS_STREAMING"));
+  curl_easy_setopt(_easy, CURLOPT_HTTP_VERSION, disable_http_2 ? CURL_HTTP_VERSION_1_1 : CURL_HTTP_VERSION_2_0);
+
+  /* keep the speed down so we don't have to buffer large amounts*/
+  curl_easy_setopt(_easy, CURLOPT_MAX_RECV_SPEED_LARGE, (curl_off_t)31415);
+  /*Add request body*/
+  if (!_body.empty()) curl_easy_setopt(_easy, CURLOPT_POSTFIELDS, _body.c_str());
+  /*Add request proxy*/
+  if (!_proxy.empty()) curl_easy_setopt(_easy, CURLOPT_PROXY, _proxy.c_str());
+
+  /*Add request headers*/
+  struct curl_slist *hdr_list = nullptr;
+  for(const auto& header : _headers) {
+    hdr_list = curl_slist_append(hdr_list, header.c_str());
+  }
+  if (hdr_list) curl_easy_setopt(_easy, CURLOPT_HTTPHEADER, hdr_list);
+
+  _status = Status_t::STATUS_AWAITING_RESTART;
+
+  _timer.expires_from_now(boost::posix_time::millisec(1));
+  _timer.async_wait(boost::bind(&AudioProducerHttp::addCurlHandle, this, boost::placeholders::_1));
+}
+void AudioProducerHttp::queueHttpGetAudio(const std::string& url, int gain, bool loop) {
+  _method = HttpMethod_t::HTTP_METHOD_GET;
+  _url = url;
+  _gain = gain;
+  _loop = loop;
+}
+void AudioProducerHttp::queueHttpPostAudio(const std::string& url, int gain, bool loop) {
+  _method = HttpMethod_t::HTTP_METHOD_POST;
+  _url = url;
+  _gain = gain;
+  _loop = loop;
+}
+void AudioProducerHttp::queueHttpPostAudio(const std::string& url, const std::string& body, std::vector<std::string>& headers, const std::string& proxy, int gain, bool loop) {
+  _method = HttpMethod_t::HTTP_METHOD_POST;
+  _url = url;
+  _body = body;
+  _headers = headers;
+  _proxy = proxy;
+  _gain = gain;
+  _loop = loop;
+}
+
+void AudioProducerHttp::addCurlHandle(const boost::system::error_code& error) {
+  if (_status == Status_t::STATUS_AWAITING_RESTART) {
+    auto rc = curl_multi_add_handle(global.multi, _easy);
+    if (mcode_test("new_conn: curl_multi_add_handle", rc) < 0) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "AudioProducerHttp::addCurlHandle: Error adding easy handle to multi handle\n");
+    }
+    else {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "AudioProducerHttp::addCurlHandle retrieving from %s\n", _url.c_str());
+    }
+  }
+}
+
+CURL* AudioProducerHttp::createEasyHandle(void) {
+  CURL* easy = curl_easy_init();
+  if(!easy) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "curl_easy_init() failed!\n");
+    return nullptr ;
+  }  
+
+  curl_easy_setopt(easy, CURLOPT_FOLLOWLOCATION, 1L);
+  curl_easy_setopt(easy, CURLOPT_USERAGENT, "jambonz/0.8.5");
+
+  // set connect timeout to 3 seconds and no total timeout as files could be large
+  curl_easy_setopt(easy, CURLOPT_CONNECTTIMEOUT_MS, 3000L);
+  curl_easy_setopt(easy, CURLOPT_TIMEOUT, 0L); // no timeout
+
+  return easy ;    
+}
+
+size_t AudioProducerHttp::static_write_cb(char* ptr, size_t size, size_t nmemb, void* userdata) {
+  return static_cast<AudioProducerHttp*>(userdata)->write_cb(ptr, size, nmemb);
+}
+
+size_t AudioProducerHttp::write_cb(void *ptr, size_t size, size_t nmemb) {
+  int8_t *data = (int8_t *) ptr;
+  size_t bytes_received = size * nmemb;
+  std::vector<int16_t> pcm_data;
+  if (_status == Status_t::STATUS_STOPPING || _status == Status_t::STATUS_STOPPED) {
+    _timer.cancel();
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, 
+    //  "AudioProducerHttp::write_cb: aborting transfer, status %s, mutex %p, buffer %p\n", status2String(_status));
+    /* this will abort the transfer */
+    return 0;
+  }
+  if (_response_code > 0 && _response_code != 200) {
+    std::string body((char *) ptr, bytes_received);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "AudioProducerHttp::write_cb: received body %s\n", body.c_str());
+    _err_msg = body;
+    _status = Status_t::STATUS_FAILED;
+    return 0;
+  }
+
+  /* throttle after reaching high water mark */
+  size_t bufSize = _buffer.size();
+  if (bufSize > BUFFER_THROTTLE_HIGH) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "AudioProducerHttp::write_cb: throttling download, buffer size is %ld\n", _buffer.size());
+    
+    // check back in 2 seconds
+    _timer.expires_from_now(boost::posix_time::millisec(2000));
+    _timer.async_wait(boost::bind(&AudioProducerHttp::throttling_cb, this, boost::placeholders::_1));
+
+    _status = Status_t::STATUS_DOWNLOAD_PAUSED;
+    return CURL_WRITEFUNC_PAUSE;
+  }
+
+  pcm_data = convert_mp3_to_linear(_mh, _gain, data, bytes_received);
+  size_t samples = pcm_data.size();
+  std::lock_guard<std::mutex> lock(_mutex); 
+
+  // Resize the buffer if necessary
+  if (_buffer.capacity() - bufSize < samples) {
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "AudioProducerHttp::write_cb growing buffer, size now %ld\n", _buffer.size()); 
+
+    //TODO: if buffer exceeds some max size, return CURL_WRITEFUNC_ERROR to abort the transfer
+    _buffer.set_capacity(_buffer.size() + std::max(samples, (size_t)BUFFER_GROW_SIZE));
+  }
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "AudioProducerHttp::write_cb: writing %ld samples to buffer\n", pcm_data.size());
+  
+  /* Push the data into the buffer */
+  _buffer.insert(_buffer.end(), pcm_data.data(), pcm_data.data() + pcm_data.size());
+  return bytes_received;
+
+}
+
+size_t AudioProducerHttp::static_header_callback(char *buffer, size_t size, size_t nitems, void* userdata) {
+  return static_cast<AudioProducerHttp*>(userdata)->header_callback(buffer, size, nitems);
+}
+
+size_t AudioProducerHttp::header_callback(char *buffer, size_t size, size_t nitems) {
+  size_t bytes_received = size * nitems;
+  const std::string prefix = "HTTP/";
+  std::string header, value;
+  std::string input(buffer, bytes_received);
+  if (parseHeader(input, header, value)) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: %s with value %s\n", header.c_str(), value.c_str());
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: %s\n", input.c_str());
+    if (input.rfind(prefix, 0) == 0) {
+      try {
+        _response_code = extract_response_code(input);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: parsed response code: %ld\n", _response_code);
+      } catch (const std::invalid_argument& e) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "header_callback: invalid response code %s\n", input.substr(prefix.length()).c_str());
+      }
+    }
+  }
+  return bytes_received;
+}
+
+void AudioProducerHttp::throttling_cb(const boost::system::error_code& error) {
+  if (_status == Status_t::STATUS_STOPPING) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "throttling_cb: session gone, resume download so we can complete\n");
+    curl_easy_pause(_easy, CURLPAUSE_CONT);
+    return;
+  }
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "throttling_cb: status is %s\n", status2String(_status));
+
+  if (!error) {
+    auto size = _buffer.size();
+
+    //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "throttling_cb: size is now %ld\n", size);
+    if (size < BUFFER_THROTTLE_LOW) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "throttling_cb: resuming download\n");
+      curl_easy_pause(_easy, CURLPAUSE_CONT);
+     return;
+    }
+
+    // check back in 2 seconds
+    _timer.expires_from_now(boost::posix_time::millisec(2000));
+    _timer.async_wait(boost::bind(&AudioProducerHttp::throttling_cb, this, boost::placeholders::_1));
+
+  } else if (125 == error.value()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "throttling_cb: timer canceled\n");
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "throttling_cb: error (%d): %s\n", error.value(), error.message().c_str());
+  }
+}
+
+int AudioProducerHttp::sock_cb(CURL *e, curl_socket_t s, int what, void *cbp, void *sockp) {
+  GlobalInfo_t *g = &global;
+
+  int *actionp = (int *) sockp;
+  static const char *whatstr[] = { "none", "IN", "OUT", "INOUT", "REMOVE"};
+
+  if(what == CURL_POLL_REMOVE) {
+    *actionp = what;
+  }
+  else {
+    if(!actionp) {
+      addsock(s, e, what, g);
+    }
+    else {
+      setsock(actionp, s, e, what, *actionp, g);
+    }
+  }
+  return 0;  
+}
+
+void AudioProducerHttp::timer_cb(const boost::system::error_code & error, GlobalInfo_t *g)
+{
+  //switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "timer_cb\n");
+
+  if(!error) {
+    CURLMcode rc = curl_multi_socket_action(g->multi, CURL_SOCKET_TIMEOUT, 0, &g->still_running);
+    mcode_test("timer_cb: curl_multi_socket_action", rc);
+    check_multi_info(g);
+  }
+}
+
+int AudioProducerHttp::multi_timer_cb(CURLM *multi, long timeout_ms, GlobalInfo_t *g) {
+
+  /* cancel running timer */
+  multi_timer.cancel();
+
+  if(timeout_ms >= 0) {
+    // from libcurl 7.88.1-10+deb12u4 does not allow call curl_multi_socket_action or curl_multi_perform in curl_multi callback directly
+    multi_timer.expires_from_now(boost::posix_time::millisec(timeout_ms ? timeout_ms : 1));
+    multi_timer.async_wait(boost::bind(&timer_cb, boost::placeholders::_1, g));
+  }
+
+  return 0;
+}
+
+curl_socket_t AudioProducerHttp::open_socket(void *clientp, curlsocktype purpose, struct curl_sockaddr *address) {
+  curl_socket_t sockfd = CURL_SOCKET_BAD;
+
+  /* restrict to IPv4 */
+  if(purpose == CURLSOCKTYPE_IPCXN && address->family == AF_INET) {
+    /* create a tcp socket object */
+    boost::asio::ip::tcp::socket *tcp_socket = new boost::asio::ip::tcp::socket(io_service);
+
+    /* open it and get the native handle*/
+    boost::system::error_code ec;
+    tcp_socket->open(boost::asio::ip::tcp::v4(), ec);
+
+    if(ec) {
+      /* An error occurred */
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't open socket [%ld][%s]\n", ec, ec.message().c_str());
+    }
+    else {
+      sockfd = tcp_socket->native_handle();
+
+      /* save it for monitoring */
+      socket_map.insert(std::pair<curl_socket_t, boost::asio::ip::tcp::socket *>(sockfd, tcp_socket));
+    }
+  }
+  return sockfd;
+}
+
+int AudioProducerHttp::close_socket(void *clientp, curl_socket_t item) {
+  std::map<curl_socket_t, boost::asio::ip::tcp::socket *>::iterator it = socket_map.find(item);
+  if(it != socket_map.end()) {
+    delete it->second;
+    socket_map.erase(it);
+  }
+  return 0;
+}
+
+int AudioProducerHttp::mcode_test(const char *where, CURLMcode code) {
+  if(CURLM_OK != code) {
+    const char *s;
+    switch(code) {
+    case CURLM_CALL_MULTI_PERFORM:
+      s = "CURLM_CALL_MULTI_PERFORM";
+      break;
+    case CURLM_BAD_HANDLE:
+      s = "CURLM_BAD_HANDLE";
+      break;
+    case CURLM_BAD_EASY_HANDLE:
+      s = "CURLM_BAD_EASY_HANDLE";
+      break;
+    case CURLM_OUT_OF_MEMORY:
+      s = "CURLM_OUT_OF_MEMORY";
+      break;
+    case CURLM_INTERNAL_ERROR:
+      s = "CURLM_INTERNAL_ERROR";
+      break;
+    case CURLM_UNKNOWN_OPTION:
+      s = "CURLM_UNKNOWN_OPTION";
+      break;
+    case CURLM_LAST:
+      s = "CURLM_LAST";
+      break;
+    default:
+      s = "CURLM_unknown";
+      break;
+    case CURLM_BAD_SOCKET:
+      s = "CURLM_BAD_SOCKET";
+      break;
+    }
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mcode_test ERROR: %s returns %s:%d\n", where, s, code);
+
+    return -1;
+  }
+  return 0 ;
+}
+
+bool AudioProducerHttp::parseHeader(const std::string& str, std::string& header, std::string& value) {
+  std::vector<std::string> parts;
+  boost::split(parts, str, boost::is_any_of(":"), boost::token_compress_on);
+
+  if (parts.size() != 2)
+      return false;
+
+  header = boost::trim_copy(parts[0]);
+  value = boost::trim_copy(parts[1]);
+  return true;
+}
+
+int AudioProducerHttp::extract_response_code(const std::string& input) {
+  std::size_t space_pos = input.find(' ');
+  if (space_pos == std::string::npos) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Invalid HTTP response format %s\n", input.c_str());
+    return 0;
+  }
+
+  std::size_t code_start_pos = space_pos + 1;
+  std::size_t code_end_pos = input.find(' ', code_start_pos);
+  if (code_end_pos == std::string::npos) {
+    code_end_pos = input.length();
+  }
+
+  std::string code_str = input.substr(code_start_pos, code_end_pos - code_start_pos);
+  int response_code = std::stoi(code_str);
+  return response_code;
+}
+
+void AudioProducerHttp::setsock(int *fdp, curl_socket_t s, CURL *e, int act, int oldact, GlobalInfo_t *g) {
+  std::map<curl_socket_t, boost::asio::ip::tcp::socket *>::iterator it = socket_map.find(s);
+
+  if(it == socket_map.end()) {
+    return;
+  }
+
+  boost::asio::ip::tcp::socket * tcp_socket = it->second;
+
+  *fdp = act;
+
+  if(act == CURL_POLL_IN) {
+    if(oldact != CURL_POLL_IN && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                  boost::bind(&event_cb, g, s,
+                                  CURL_POLL_IN, boost::placeholders::_1, fdp));
+    }
+  }
+  else if(act == CURL_POLL_OUT) {
+    if(oldact != CURL_POLL_OUT && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                    CURL_POLL_OUT, boost::placeholders::_1, fdp));
+    }
+  }
+  else if(act == CURL_POLL_INOUT) {
+    if(oldact != CURL_POLL_IN && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                  boost::bind(&event_cb, g, s,
+                                  CURL_POLL_IN, boost::placeholders::_1, fdp));
+    }
+    if(oldact != CURL_POLL_OUT && oldact != CURL_POLL_INOUT) {
+      tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                    CURL_POLL_OUT, boost::placeholders::_1, fdp));
+    }
+  }
+}
+
+/* Called by asio when there is an action on a socket */
+void AudioProducerHttp::event_cb(GlobalInfo_t *g, curl_socket_t s, int action, const boost::system::error_code & error, int *fdp) {
+  int f = *fdp;
+
+
+  // Socket already POOL REMOVED.
+  if (f == CURL_POLL_REMOVE) {
+    remsock(fdp, g);
+    return;
+  }
+
+  if(socket_map.find(s) == socket_map.end()) {
+    return;
+  }
+
+  /* make sure the event matches what are wanted */
+  if(f == action || f == CURL_POLL_INOUT) {
+    if(error) {
+      action = CURL_CSELECT_ERR;
+    }
+    CURLMcode rc = curl_multi_socket_action(g->multi, s, action, &g->still_running);
+
+    mcode_test("event_cb: curl_multi_socket_action", rc);
+    check_multi_info(g);
+
+    if(g->still_running <= 0) {
+      multi_timer.cancel();
+    }
+
+    /* keep on watching.
+      * the socket may have been closed and/or fdp may have been changed
+      * in curl_multi_socket_action(), so check them both */
+    if(!error && socket_map.find(s) != socket_map.end() &&
+        (f == action || f == CURL_POLL_INOUT)) {
+      boost::asio::ip::tcp::socket *tcp_socket = socket_map.find(s)->second;
+
+      if(action == CURL_POLL_IN) {
+        tcp_socket->async_read_some(boost::asio::null_buffers(),
+                                    boost::bind(&event_cb, g, s,
+                                                action, boost::placeholders::_1, fdp));
+      }
+      if(action == CURL_POLL_OUT) {
+        tcp_socket->async_write_some(boost::asio::null_buffers(),
+                                      boost::bind(&event_cb, g, s,
+                                                  action, boost::placeholders::_1, fdp));
+      } 
+    }
+  }
+}
+
+void AudioProducerHttp::remsock(int *f, GlobalInfo_t *g) {
+  if(f) {
+    free(f);
+    f = NULL;
+  }
+}
+
+void AudioProducerHttp::addsock(curl_socket_t s, CURL *easy, int action, GlobalInfo_t *g) {
+  /* fdp is used to store current action */
+  int *fdp = (int *) calloc(sizeof(int), 1);
+
+  setsock(fdp, s, easy, action, 0, g);
+  curl_multi_assign(g->multi, s, fdp);
+}
+
+/* Check for completed transfers, and remove their easy handles */
+void AudioProducerHttp::check_multi_info(GlobalInfo_t *g) {
+  CURLMsg *msg;
+  int msgs_left;
+  AudioProducerHttp *ap;
+  CURL *easy;
+  CURLcode res;
+  
+  while((msg = curl_multi_info_read(g->multi, &msgs_left))) {
+    if(msg->msg == CURLMSG_DONE) {
+      long response_code;
+      double namelookup=0, connect=0, total=0 ;
+      char *ct = NULL ;
+
+      easy = msg->easy_handle;
+      res = msg->data.result;
+      curl_easy_getinfo(easy, CURLINFO_PRIVATE, &ap);
+      curl_easy_getinfo(easy, CURLINFO_RESPONSE_CODE, &response_code);
+      curl_easy_getinfo(easy, CURLINFO_CONTENT_TYPE, &ct);
+
+      curl_easy_getinfo(easy, CURLINFO_NAMELOOKUP_TIME, &namelookup);
+      curl_easy_getinfo(easy, CURLINFO_CONNECT_TIME, &connect);
+      curl_easy_getinfo(easy, CURLINFO_TOTAL_TIME, &total);
+
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "curl done, response code %d, status %s\n", response_code, status2String(ap->getStatus()));
+
+      bool restart = ap->isLoopedAudio() && ap->getStatus() != Status_t::STATUS_STOPPING && response_code == 200;
+      if (restart) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "restarting looped audio\n");
+        ap->getTimer().expires_from_now(boost::posix_time::millisec(1000));
+        ap->getTimer().async_wait(boost::bind(&AudioProducerHttp::static_restart_cb, boost::placeholders::_1, ap));
+      }
+      else {
+        ap->cleanup(Status_t::STATUS_DOWNLOAD_COMPLETE, (int) response_code);
+      }
+    }
+  }
+}
+
+void AudioProducerHttp::static_restart_cb(const boost::system::error_code& error, void* userdata) {
+  static_cast<AudioProducerHttp*>(userdata)->restart_cb(error);
+}
+
+void AudioProducerHttp::restart_cb(const boost::system::error_code& error) {
+  if (_status == Status_t::STATUS_STOPPING) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "restart_cb: session gone\n");
+    return;
+  }
+  if (!error) {
+    reset();
+    start(_callback);
+  }
+  else if (error.value() == boost::asio::error::operation_aborted) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "restart_cb: %s, cancelling retrieve\n", error.message().c_str());
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "restart_cb: error (%d): %s\n", error.value(), error.message().c_str());
+  }
+}
+
+
+void AudioProducerHttp::stop() {
+  cleanup(Status_t::STATUS_STOPPED, 0);
+  return;
+}
+
+void AudioProducerHttp::reset() {
+  {
+    std::lock_guard<std::mutex> lock(_mutex);
+    if (_easy) {
+      curl_multi_remove_handle(global.multi, _easy);
+      curl_easy_cleanup(_easy);
+      _easy = nullptr;
+    }
+    if (_mh) {
+      mpg123_close(_mh);
+      mpg123_delete(_mh);
+      _mh = nullptr;
+    }
+  }
+  _err_msg.clear();
+  _response_code = 0;
+  _timer.cancel();
+  _status = Status_t::STATUS_NONE;
+}
+
+void AudioProducerHttp::cleanup(Status_t status, int response_code) {
+  std::string errMsg = response_code > 200 ? "http response: " + std::to_string(response_code) : "";
+  reset();
+  _status = status;
+  notifyDone(!errMsg.empty(), errMsg);
+}

mod_dub/ap_http.h

@@ -0,0 +1,135 @@
+#ifndef __AP_HTTP_H__
+#define __AP_HTTP_H__
+
+#include "ap.h"
+#include <curl/curl.h>
+#include <mpg123.h>
+#include <boost/asio.hpp>
+
+
+typedef struct
+{
+  CURLM *multi;
+  int still_running;
+} GlobalInfo_t;
+
+class AudioProducerHttp : public AudioProducer {
+public:
+
+  typedef enum
+  {
+    STATUS_NONE = 0,
+    STATUS_FAILED,
+    STATUS_DOWNLOAD_IN_PROGRESS,
+    STATUS_DOWNLOAD_PAUSED,
+    STATUS_DOWNLOAD_COMPLETE,
+    STATUS_AWAITING_RESTART,
+    STATUS_STOPPING,
+    STATUS_STOPPED
+  } Status_t;
+
+  typedef enum {
+    HTTP_METHOD_GET = 0,
+    HTTP_METHOD_POST
+  } HttpMethod_t;
+
+  static const char* status2String(Status_t status) {
+    static const char* statusStrings[] = {
+      "STATUS_NONE",
+      "STATUS_FAILED",
+      "STATUS_DOWNLOAD_IN_PROGRESS",
+      "STATUS_DOWNLOAD_PAUSED",
+      "STATUS_DOWNLOAD_COMPLETE",
+      "STATUS_AWAITING_RESTART",
+      "STATUS_STOPPED"
+    };
+
+    if (status >= 0 && status < sizeof(statusStrings) / sizeof(statusStrings[0]))
+    {
+      return statusStrings[status];
+    }
+    else
+    {
+      return "UNKNOWN_STATUS";
+    }
+  }
+  AudioProducerHttp(
+    std::mutex& mutex,
+    CircularBuffer_t& circularBuffer,
+    int sampleRate
+  );
+  virtual ~AudioProducerHttp();
+
+  virtual void start(std::function<void(bool, const std::string&)> callback);
+  void addCurlHandle(const boost::system::error_code& error);
+  virtual void stop();
+  void cleanup(Status_t status, int response_code);
+  void reset();
+
+  void queueHttpGetAudio(const std::string& url, int gain = 0, bool loop = false);
+  void queueHttpPostAudio(const std::string& url, int gain = 0, bool loop = false);
+  void queueHttpPostAudio(const std::string& url, const std::string& body, std::vector<std::string>& headers, const std::string& proxy, int gain = 0, bool loop = false);
+
+  Status_t getStatus() const { return _status; }
+  void setStatus(Status_t status) { _status = status; }
+
+  boost::asio::deadline_timer& getTimer() { return _timer; }
+  
+  static bool initialized;
+  static std::thread worker_thread;
+  static boost::asio::io_service io_service;
+  static void threadFunc();
+  static std::map<curl_socket_t, boost::asio::ip::tcp::socket *> socket_map;
+  static boost::asio::deadline_timer multi_timer;
+
+
+  static GlobalInfo_t global;
+
+  static int sock_cb(CURL *e, curl_socket_t s, int what, void *cbp, void *sockp);
+  static int multi_timer_cb(CURLM *multi, long timeout_ms, GlobalInfo_t *g);
+  static void timer_cb(const boost::system::error_code & error, GlobalInfo_t *g);
+  static void addsock(curl_socket_t s, CURL *easy, int action, GlobalInfo_t *g);
+  static void setsock(int *fdp, curl_socket_t s, CURL *e, int act, int oldact, GlobalInfo_t *g);
+  static int mcode_test(const char *where, CURLMcode code);
+  static void check_multi_info(GlobalInfo_t *g);
+  static void event_cb(GlobalInfo_t *g, curl_socket_t s, int action, const boost::system::error_code & error, int *fdp);
+  static void remsock(int *f, GlobalInfo_t *g);
+  
+  static size_t static_write_cb(char* ptr, size_t size, size_t nmemb, void* userdata);
+  size_t write_cb(void *ptr, size_t size, size_t nmemb);
+
+  static size_t static_header_callback(char *buffer, size_t size, size_t nitems, void* userdata);
+  size_t header_callback(char *buffer, size_t size, size_t nitems);
+
+  void throttling_cb(const boost::system::error_code& error);
+
+  static void static_restart_cb(const boost::system::error_code& error, void* userdata);
+  void restart_cb(const boost::system::error_code& error);
+
+  static int close_socket(void *clientp, curl_socket_t item);
+  static curl_socket_t open_socket(void *clientp, curlsocktype purpose, struct curl_sockaddr *address);
+
+private:
+
+  static void _init();
+  static void _deinit();
+
+  CURL* createEasyHandle();
+  bool parseHeader(const std::string& str, std::string& header, std::string& value);
+  int extract_response_code(const std::string& input);
+
+  HttpMethod_t _method;
+  std::string _url;
+  std::string _body;
+  std::string _proxy;
+  std::vector<std::string> _headers;
+  Status_t _status;
+  mpg123_handle *_mh;
+  CURL *_easy;
+  char _error[CURL_ERROR_SIZE]; // curl error buffer
+  std::string _err_msg;
+  int _response_code;
+  boost::asio::deadline_timer _timer;
+};
+
+#endif

mod_dub/common.h

@@ -0,0 +1,9 @@
+#ifndef _COMMON_H_
+#define _COMMON_H_
+
+#include <boost/circular_buffer.hpp>
+
+typedef boost::circular_buffer<int16_t> CircularBuffer_t;
+
+
+#endif

mod_dub/dub_glue.cpp

@@ -0,0 +1,209 @@
+#include "mod_dub.h"
+#include "tts_vendor_parser.h"
+#include "track.h"
+#include "vector_math.h"
+#include <string>
+#include <queue>
+
+#include <switch.h>
+
+#include <curl/curl.h>
+
+#include <boost/circular_buffer.hpp>
+
+typedef boost::circular_buffer<int16_t> CircularBuffer_t;
+#define INIT_BUFFER_SIZE (80000)
+
+extern "C" {
+
+  Track* find_track_by_name(void** tracks, const std::string& trackName) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "find_track_by_name: searching for %s\n", trackName.c_str());
+
+    for (int i = 0; i < MAX_DUB_TRACKS; i++) {
+      Track* track = static_cast<Track*>(tracks[i]);
+      std::string name = track ? track->getTrackName() : "null";
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "find_track_by_name: offset %d: %s\n", i, name.c_str());
+      if (track && 0 == track->getTrackName().compare(trackName)) {
+        return track;
+      }
+    }
+    return nullptr;
+  }
+
+  switch_status_t add_track(struct cap_cb* cb, char* trackName, int sampleRate) {
+    Track* existingTrack = find_track_by_name(cb->tracks, trackName);
+    if (existingTrack) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "add_track: track %s already exists\n", trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+
+    for (int i = 0; i < MAX_DUB_TRACKS; i++) {
+      if (!cb->tracks[i]) {
+        cb->tracks[i] = new Track(trackName, sampleRate);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "add_track: added track %s at offset %d\n", trackName, i);
+        return SWITCH_STATUS_SUCCESS;
+      }
+    }
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "add_track: no room for track %s\n", trackName);
+    return SWITCH_STATUS_FALSE;
+  }
+
+  switch_status_t silence_dub_track(struct cap_cb* cb, char* trackName) {
+    Track* track = find_track_by_name(cb->tracks, trackName);
+    if (track) {
+      track->removeAllAudio();
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "silence_dub_track: silenced track %s\n", trackName);
+      return SWITCH_STATUS_SUCCESS;
+    }
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "silence_dub_track: track %s not found\n", trackName);
+    return SWITCH_STATUS_FALSE;
+  }
+
+  switch_status_t remove_dub_track(struct cap_cb* cb, char* trackName) {
+    for (int i = 0; i < MAX_DUB_TRACKS; i++) {
+      Track* track = static_cast<Track*>(cb->tracks[i]);
+      if (track && track->getTrackName() == trackName) {
+        track->removeAllAudio();
+        delete track;
+        cb->tracks[i] = nullptr;
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "remove_dub_track: removed track %s\n", trackName);
+        return SWITCH_STATUS_SUCCESS;
+      }
+    }
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "remove_dub_track: track %s not found\n", trackName);
+    return SWITCH_STATUS_FALSE;
+  }
+
+  switch_status_t play_dub_track(struct cap_cb* cb, char* trackName, char* url, int loop, int gain) {
+    bool isHttp = strncmp(url, "http", 4) == 0;
+    Track* track = find_track_by_name(cb->tracks, trackName);
+
+    if (!track) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "play_dub_track: track %s not found\n", trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+
+    if (isHttp) {
+      track->queueHttpGetAudio(url, gain, loop);
+    }
+    else {
+      track->queueFileAudio(url, gain, loop);
+    }
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t say_dub_track(struct cap_cb* cb, char* trackName, char* text, int gain) {
+    std::vector<std::string> headers;
+    std::string url, body, proxy;
+    Track* track = find_track_by_name(cb->tracks, trackName);
+
+    if (!track) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "play_dub_track: track %s not found\n", trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+    if (tts_vendor_parse_text(text, url, body, headers, proxy) != SWITCH_STATUS_SUCCESS) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "say_dub_track: failed to parse text\n");
+      return SWITCH_STATUS_FALSE;
+    }
+    track->queueHttpPostAudio(url, body, headers, proxy, gain);
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+
+  /* module load and unload */
+  switch_status_t dub_init() {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t dub_cleanup() {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t dub_session_cleanup(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug) {
+    switch_channel_t *channel = switch_core_session_get_channel(session);
+
+    if (bug) {
+      struct cap_cb *cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+      switch_mutex_lock(cb->mutex);
+
+      if (!switch_channel_get_private(channel, MY_BUG_NAME)) {
+        // race condition
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "%s Bug is not attached (race).\n", switch_channel_get_name(channel));
+        switch_mutex_unlock(cb->mutex);
+        return SWITCH_STATUS_FALSE;
+      }
+      switch_channel_set_private(channel, MY_BUG_NAME, NULL);
+
+      for (int i = 0; i < MAX_DUB_TRACKS; i++) {
+        Track* track = static_cast<Track*>(cb->tracks[i]);
+        if (track) {
+          track->removeAllAudio();
+          delete track;
+          cb->tracks[i] = nullptr;
+        }
+      }
+
+      if (!channelIsClosing) {
+        switch_core_media_bug_remove(session, &bug);
+      }
+
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "dub_session_cleanup: removed bug and cleared tracks\n");
+			switch_mutex_unlock(cb->mutex);
+    }
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s dub_session_cleanup: Bug is not attached.\n", switch_channel_get_name(channel));
+    return SWITCH_STATUS_FALSE;
+  }
+
+  switch_bool_t dub_speech_frame(switch_media_bug_t *bug, void* user_data) {
+    switch_core_session_t *session = switch_core_media_bug_get_session(bug);
+    struct cap_cb *cb = (struct cap_cb *) user_data;
+
+    if (switch_mutex_trylock(cb->mutex) == SWITCH_STATUS_SUCCESS) {
+
+      /* check if any tracks have audio to contribute */
+      std::vector<Track*> activeTracks;
+      activeTracks.reserve(MAX_DUB_TRACKS);
+      for (int i = 0; i < MAX_DUB_TRACKS; i++) {
+        if (cb->tracks[i]) {
+          auto track = static_cast<Track*>(cb->tracks[i]);
+          if (track->hasAudio_NoLock()) activeTracks.push_back(static_cast<Track*>(cb->tracks[i]));
+        }
+      }
+
+      if (activeTracks.size() == 0 && cb->gain == 0) {
+        switch_mutex_unlock(cb->mutex);
+        return SWITCH_TRUE;
+      }
+
+      switch_frame_t* rframe = switch_core_media_bug_get_write_replace_frame(bug);
+      int16_t *fp = reinterpret_cast<int16_t*>(rframe->data);
+
+      rframe->channels = 1;
+      rframe->datalen = rframe->samples * sizeof(int16_t);
+
+      /* apply gain to audio in main channel if requested*/
+      if (cb->gain != 0) {
+        vector_change_sln_volume_granular(fp, rframe->samples, cb->gain);
+      }
+
+      /* now mux in the data from tracks */
+      for (auto track : activeTracks) {
+        int16_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
+        memset(data, 0, sizeof(data));
+        auto samples = track->retrieveAndClearAudio(data, rframe->samples);
+        if (samples > 0) {
+          vector_add(fp, data, rframe->samples);
+        }
+      }
+      vector_normalize(fp, rframe->samples);
+
+      switch_core_media_bug_set_write_replace_frame(bug, rframe);
+      switch_mutex_unlock(cb->mutex);
+    }
+    return SWITCH_TRUE;
+  }
+}

mod_dub/dub_glue.h

@@ -0,0 +1,16 @@
+#ifndef __DUB_GLUE_H__
+#define __DUB_GLUE_H__
+
+switch_status_t dub_init();
+switch_status_t dub_cleanup();
+
+switch_status_t add_track(struct cap_cb* cb, char* trackName, int sampleRate);
+switch_status_t silence_dub_track(struct cap_cb* cb, char* trackName);
+switch_status_t remove_dub_track(struct cap_cb* cb, char* trackName);
+switch_status_t play_dub_track(struct cap_cb* cb, char* trackName, char* url, int loop, int gain);
+switch_status_t say_dub_track(struct cap_cb* cb, char* trackName, char* text, int gain);
+
+switch_status_t dub_session_cleanup(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug);
+switch_bool_t dub_speech_frame(switch_media_bug_t *bug, void* user_data);
+
+#endif

mod_dub/mod_dub.c

@@ -0,0 +1,333 @@
+/* 
+ *
+ * mod_dub.c
+ *
+ */
+#include "mod_dub.h"
+#include <stdlib.h>
+#include <switch.h>
+#include <switch_curl.h>
+#include "dub_glue.h"
+#include <string.h>
+
+/* Prototypes */
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_dub_shutdown);
+SWITCH_MODULE_LOAD_FUNCTION(mod_dub_load);
+
+SWITCH_MODULE_DEFINITION(mod_dub, mod_dub_load, mod_dub_shutdown, NULL);
+
+static switch_bool_t capture_callback(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type)
+{
+	switch_core_session_t *session = switch_core_media_bug_get_session(bug);
+  switch_bool_t ret = SWITCH_TRUE;
+
+	switch (type) {
+	case SWITCH_ABC_TYPE_INIT:
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Got SWITCH_ABC_TYPE_INIT.\n");
+		break;
+
+	case SWITCH_ABC_TYPE_CLOSE:
+		{
+			dub_session_cleanup(session, 1, bug);
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Finished SWITCH_ABC_TYPE_CLOSE.\n");
+		}
+		break;
+	
+	case SWITCH_ABC_TYPE_WRITE_REPLACE:
+		ret = dub_speech_frame(bug, user_data);
+		break;
+
+	default:
+		break;
+	}
+
+	return ret;
+}
+
+static switch_status_t dub_set_gain(switch_core_session_t *session, int gain) {
+  switch_channel_t *channel = switch_core_session_get_channel(session);
+  switch_media_bug_t *bug = NULL;
+  struct cap_cb *cb = NULL;
+
+  if (switch_channel_pre_answer(channel) != SWITCH_STATUS_SUCCESS) {
+		return SWITCH_STATUS_FALSE;
+	}
+
+  if (!(bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME))) {
+    cb =(struct cap_cb *) switch_core_session_alloc(session, sizeof(struct cap_cb));
+    memset(cb, 0, sizeof(struct cap_cb));
+    switch_mutex_init(&cb->mutex, SWITCH_MUTEX_NESTED, switch_core_session_get_pool(session));    
+  }
+  else {
+    cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+  }
+  cb->gain = gain;
+
+  /* check again under lock */
+  switch_mutex_lock(cb->mutex);
+  if (!(bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME))) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "dub_set_gain: adding bug so we can set gain on main channel\n");
+    if (switch_core_media_bug_add(session, MY_BUG_NAME, NULL, capture_callback, (void *) cb, 0, SMBF_WRITE_REPLACE, &bug) != SWITCH_STATUS_SUCCESS) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_set_gain: error adding bug!\n");
+      switch_mutex_unlock(cb->mutex);
+      return SWITCH_STATUS_FALSE;
+    }
+    switch_channel_set_private(channel, MY_BUG_NAME, bug);
+  }
+  else {
+    cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+    cb->gain = gain;
+  }
+  switch_mutex_unlock(cb->mutex);
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "dub_set_gain: setting gain to %d\n", gain);
+  return SWITCH_STATUS_SUCCESS;
+}
+
+
+static switch_status_t dub_add_track(switch_core_session_t *session, char* trackName) {
+  switch_channel_t *channel = switch_core_session_get_channel(session);
+  switch_media_bug_t *bug = NULL;
+  struct cap_cb *cb = NULL;
+  int samples_per_second;
+
+  switch_codec_implementation_t write_impl = { 0 };
+
+	if (switch_channel_pre_answer(channel) != SWITCH_STATUS_SUCCESS) {
+		return SWITCH_STATUS_FALSE;
+	}
+
+  if (!(bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME))) {
+    /* allocate per-session memory and use track 0 */
+    cb =(struct cap_cb *) switch_core_session_alloc(session, sizeof(struct cap_cb));
+    memset(cb, 0, sizeof(struct cap_cb));
+    switch_mutex_init(&cb->mutex, SWITCH_MUTEX_NESTED, switch_core_session_get_pool(session));    
+  }
+  else {
+    /* retrieve the bug and search for an empty track */
+    cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+  }
+
+  switch_core_session_get_write_impl(session, &write_impl);
+	samples_per_second = !strcasecmp(write_impl.iananame, "g722") ? write_impl.actual_samples_per_second : write_impl.samples_per_second;
+
+  switch_mutex_lock(cb->mutex);
+  if (add_track(cb, trackName, samples_per_second) != SWITCH_STATUS_SUCCESS) {
+    switch_mutex_unlock(cb->mutex);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_add_track: error adding track %s\n", trackName);
+    return SWITCH_STATUS_FALSE;
+  }
+
+  if (!bug) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "dub_add_track: adding bug for track %s\n", trackName);
+    if (switch_core_media_bug_add(session, MY_BUG_NAME, NULL, capture_callback, (void *) cb, 0, SMBF_WRITE_REPLACE, &bug) != SWITCH_STATUS_SUCCESS) {
+      switch_mutex_unlock(cb->mutex);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_add_track: error adding bug!\n");
+      return SWITCH_STATUS_FALSE;
+    }
+    switch_channel_set_private(channel, MY_BUG_NAME, bug);
+  }
+  switch_mutex_unlock(cb->mutex);
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+static switch_status_t dub_remove_track(switch_core_session_t *session, char* trackName) {
+  switch_channel_t *channel = switch_core_session_get_channel(session);
+  switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME);
+  switch_status_t status = SWITCH_STATUS_FALSE;
+  
+  if (bug) {
+    struct cap_cb *cb =(struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+
+    switch_mutex_lock(cb->mutex);
+    status = remove_dub_track(cb, trackName);
+    switch_mutex_unlock(cb->mutex);
+    if (status != SWITCH_STATUS_SUCCESS) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_remove_track: error removing track %s\n", trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_remove_track: bug not found\n");
+  }
+
+  return status;
+}
+
+static switch_status_t dub_silence_track(switch_core_session_t *session, char* trackName) {
+  switch_channel_t *channel = switch_core_session_get_channel(session);
+  switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME);
+  switch_status_t status = SWITCH_STATUS_FALSE;
+
+  // DH: I found it is much simpler to implement silence as a sequence of remove and add operations  
+  if (bug) {
+    struct cap_cb *cb =(struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+    status = remove_dub_track(cb, trackName);
+    if (status != SWITCH_STATUS_SUCCESS) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_silence_track: error finding track %s\n", trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+    status = dub_add_track(session, trackName);
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_silence_track: bug not found\n");
+  }
+
+  return status;
+}
+
+static switch_status_t dub_play_on_track(switch_core_session_t *session, char* trackName, char* url, int loop, int gain) {
+  switch_channel_t *channel = switch_core_session_get_channel(session);
+  switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME);
+  switch_status_t status = SWITCH_STATUS_FALSE;
+  
+  if (bug) {
+    struct cap_cb *cb =(struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+    status = play_dub_track(cb, trackName, url, loop, gain);
+    if (status != SWITCH_STATUS_SUCCESS) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_play_on_track: error playing %s on track %s\n", url, trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+  }
+  return status;
+}
+
+static switch_status_t dub_say_on_track(switch_core_session_t *session, char* trackName, char* text, int gain) {
+  switch_channel_t *channel = switch_core_session_get_channel(session);
+  switch_media_bug_t *bug = (switch_media_bug_t*) switch_channel_get_private(channel, MY_BUG_NAME);
+  switch_status_t status = SWITCH_STATUS_FALSE;
+  
+  if (bug) {
+    struct cap_cb *cb =(struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+    status = say_dub_track(cb, trackName, text, gain);
+    if (status != SWITCH_STATUS_SUCCESS) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_say_on_track: error saying %s on track %s\n", text, trackName);
+      return SWITCH_STATUS_FALSE;
+    }
+  }
+  return status;
+}
+
+#define DUB_API_SYNTAX "<uuid> [addTrack|removeTrack|silenceTrack|playOnTrack|sayOnTrack|setGain] track [url|text|gain] [gain] [loop]"
+#define MAX_PARAMS 6
+SWITCH_STANDARD_API(dub_function)
+{
+	char *mycmd = NULL, *argv[MAX_PARAMS] = { 0 };
+	int argc = 0;
+  int error_written = 0;
+	switch_status_t status = SWITCH_STATUS_FALSE;
+
+	if (!zstr(cmd) && (mycmd = strdup(cmd))) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "dub_function: %s\n", mycmd);
+    argc = switch_separate_string(mycmd, ' ', argv, (sizeof(argv) / sizeof(argv[0])));
+	}
+
+  if (zstr(cmd) || argc < 3 || zstr(argv[0]) || zstr(argv[1]) || zstr(argv[2])) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error with command %s.\n", cmd);
+  }
+  else {
+    switch_core_session_t *session = NULL;
+		if ((session = switch_core_session_locate(argv[0]))) {
+      char* action = argv[1];
+      char* track = argv[2];
+
+      if (0 == strcmp(action, "setGain")) {
+        int gain = atoi(track);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "setGain %d\n", gain);
+        status = dub_set_gain(session, gain);
+      }
+      else if (0 == strcmp(action, "addTrack")) {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "addTrack %s\n", track);
+        status = dub_add_track(session, track);
+      }
+      else if (0 == strcmp(action, "removeTrack")) {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "removeTrack %s\n", track);
+        status = dub_remove_track(session, track);
+      }
+      else if (0 == strcmp(action, "silenceTrack")) {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "silenceTrack %s\n", track);
+        status = dub_silence_track(session, track);
+      }
+      else if (0 == strcmp(action, "playOnTrack")) {
+        if (argc < 4) {
+          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "playOnTrack requires a url\n");
+          stream->write_function(stream, "-USAGE: %s\n", DUB_API_SYNTAX);
+          error_written = 1;
+        }
+        else {
+          char* url = argv[3];
+          int loop = argc > 4 && 0 == strcmp(argv[4], "loop");
+          int gain = argc > 5 ? atoi(argv[5]) : 0;
+          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, 
+            "playOnTrack %s, %s gain %d\n", url, loop ? "loop" : "once", gain);
+          status = dub_play_on_track(session, track, url, loop, gain);
+        }
+      }
+      else if (0 == strcmp(action, "sayOnTrack")) {
+        if (argc < 4) {
+          stream->write_function(stream, "-USAGE: %s\n", DUB_API_SYNTAX);
+          error_written = 1;
+        }
+        else {
+          char* text = argv[3];
+          int gain = argc > 5 ? atoi(argv[5]) : 0;
+          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "sayOnTrack %s gain %d\n", text, gain);
+          status = dub_say_on_track(session, track, text, gain);
+        }
+      }
+      else {
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error with command %s %s %s.\n", cmd, argv[0], argv[1]);
+        stream->write_function(stream, "-USAGE: %s\n", DUB_API_SYNTAX);
+          error_written = 1;
+      }
+      switch_core_session_rwunlock(session);
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error locating session for %s.\n", cmd);
+      stream->write_function(stream, "-ERR Invalid session\n");
+      error_written = 1;
+    }
+  }
+
+	if (status == SWITCH_STATUS_SUCCESS) {
+		stream->write_function(stream, "+OK Success\n");
+	} else if (!error_written){
+		stream->write_function(stream, "-ERR Operation Failed\n");
+	}
+
+	switch_safe_free(mycmd);
+	return SWITCH_STATUS_SUCCESS;
+}
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_dub_load)
+{
+	switch_api_interface_t *api_interface;
+
+	*module_interface = switch_loadable_module_create_module_interface(pool, modname);
+
+	dub_init();
+
+	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_dub loaded\n");
+
+	SWITCH_ADD_API(api_interface, "uuid_dub", "dub mp3 track over channel audio", dub_function, DUB_API_SYNTAX);
+	switch_console_set_complete("add uuid_dub addTrack <trackname>");
+	switch_console_set_complete("add uuid_dub removeTrack <trackname>");
+	switch_console_set_complete("add uuid_dub silenceTrack <trackname>");
+	switch_console_set_complete("add uuid_dub playOnTrack <trackname> <url|file> [loop|once] [gain]");
+	switch_console_set_complete("add uuid_dub setGain <gain>");
+	switch_console_set_complete("add uuid_dub stop ");
+
+	/* indicate that the module should continue to be loaded */
+	return SWITCH_STATUS_SUCCESS;
+}
+
+/*
+  Called when the system shuts down
+  Macro expands to: switch_status_t mod_dub_shutdown() */
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_dub_shutdown)
+{
+	dub_cleanup();
+	return SWITCH_STATUS_SUCCESS;
+}
+

mod_dub/mod_dub.h

@@ -0,0 +1,28 @@
+#ifndef __MOD_DUB_H__
+#define __MOD_DUB_H__
+
+#include <switch.h>
+#include <speex/speex_resampler.h>
+
+#include <mpg123.h>
+
+#include <unistd.h>
+
+#define MY_BUG_NAME "_dub_"
+#define MAX_SESSION_ID (256)
+#define MAX_BUG_LEN (64)
+#define MAX_URL_LEN (1024)
+#define MAX_DUB_TRACKS (2)
+
+/* per-channel data */
+typedef void (*responseHandler_t)(switch_core_session_t* session, const char* json, const char* bugname, const char* details);
+
+
+struct cap_cb {
+	switch_mutex_t *mutex;
+  int gain;
+  void *tracks[MAX_DUB_TRACKS];
+};
+
+
+#endif

mod_dub/mpg_decode.cpp

@@ -0,0 +1,58 @@
+
+#include "mpg_decode.h"
+#include "vector_math.h"
+
+std::vector<int16_t> convert_mp3_to_linear(mpg123_handle *mh, int gain, int8_t *data, size_t len) {
+  std::vector<int16_t> linear_data;
+  int eof = 0;
+  int mp3err = 0;
+
+  if(mpg123_feed(mh, (const unsigned char*) data, len) == MPG123_OK) {
+    while(!eof) {
+      size_t usedlen = 0;
+      off_t frame_offset;
+      unsigned char* audio;
+
+      int decode_status = mpg123_decode_frame(mh, &frame_offset, &audio, &usedlen);
+
+      switch(decode_status) {
+        case MPG123_NEW_FORMAT:
+          continue;
+
+        case MPG123_OK:
+          {
+            size_t samples = usedlen / sizeof(int16_t);
+            linear_data.insert(linear_data.end(), reinterpret_cast<int16_t*>(audio), reinterpret_cast<int16_t*>(audio) + samples);
+          }
+          break;
+
+        case MPG123_DONE:
+        case MPG123_NEED_MORE:
+          eof = 1;
+          break;
+
+        case MPG123_ERR:
+        default:
+          if(++mp3err >= 5) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Decoder Error!\n");
+            eof = 1;
+          }
+      }
+
+      if (eof)
+        break;
+
+      mp3err = 0;
+    }
+
+    if (gain != 0) {
+      vector_change_sln_volume_granular(linear_data.data(), linear_data.size(), gain);
+    }
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error feeding data to mpg123\n");
+  }
+
+
+  return linear_data;
+}

mod_dub/mpg_decode.h

@@ -0,0 +1,11 @@
+
+#ifndef MPG_DECODE_H
+#define MPG_DECODE_H
+
+#include <vector>
+#include <mpg123.h>
+#include "switch.h"
+
+std::vector<int16_t> convert_mp3_to_linear(mpg123_handle *mh, int gain, int8_t *data, size_t len);
+
+#endif

mod_dub/track.cpp

@@ -0,0 +1,157 @@
+#include "track.h"
+#include "ap_file.h"
+#include "ap_http.h"
+#include "switch.h"
+
+#define INIT_BUFFER_SIZE (80000)
+
+Track::Track(const std::string& trackName, int sampleRate) : _trackName(trackName), _sampleRate(sampleRate), 
+  _buffer(INIT_BUFFER_SIZE), _stopping(false)
+{
+}
+
+Track::~Track() {
+  removeAllAudio();
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Track::~Track: track %s\n", _trackName.c_str());
+}
+
+/**
+ * @brief called when an audio producer has finished retrieving the audio.
+ * If we have another producer queued, then start it.
+ * 
+ * @param hasError 
+ * @param errMsg 
+ */
+void Track::onPlayDone(bool hasError, const std::string& errMsg) {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Track::onPlayDone %s\n", _trackName.c_str());
+
+  if (!_stopping) {
+    std::shared_ptr<AudioProducer> apDone, apNext;  // to retain a ref so the ap is not destroyed while under lock below -- avoid deadlock
+  
+    if (hasError) switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "onPlayDone: error: %s\n", errMsg.c_str());
+    {
+      std::lock_guard<std::mutex> lock(_mutex);
+      if (!_apQueue.empty()) {
+        apDone = _apQueue.front();
+        _apQueue.pop();
+      }
+       if (!_apQueue.empty()) apNext = _apQueue.front();      
+    }
+    if (apNext) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "onPlayDone: starting queued audio on track %s\n", _trackName.c_str());
+      apNext->start(std::bind(&Track::onPlayDone, this, std::placeholders::_1, std::placeholders::_2));
+    }
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "onPlayDone: track %s stopping\n", _trackName.c_str());
+  }
+}
+
+void Track::queueFileAudio(const std::string& path, int gain, bool loop) {
+  bool startIt = false;
+  if (_stopping) return;
+
+  auto ap = std::make_shared<AudioProducerFile>(_mutex, _buffer, _sampleRate);
+  ap->queueFileAudio(path, gain, loop);
+  {
+    std::lock_guard<std::mutex> lock(_mutex);
+    _apQueue.push(ap);
+    startIt = _apQueue.size() == 1;
+  }
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, 
+    "queueFileAudio: queue file %s on track %s\n", _trackName.c_str(), path.c_str());
+
+  if (startIt) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, 
+      "queueFileAudio: track %s starting queued file as none are in progress\n", _trackName.c_str());
+    try {
+      ap->start(std::bind(&Track::onPlayDone, this, std::placeholders::_1, std::placeholders::_2));
+    } catch (std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, 
+        "queueFileAudio: track %s error starting %s: %s\n", _trackName.c_str(), path.c_str(), e.what());
+      onPlayDone(true, e.what());
+    }
+  }
+}
+
+void Track::queueHttpGetAudio(const std::string& url, int gain, bool loop) {
+  bool startIt = false;
+  if (_stopping) return;
+  auto ap = std::make_shared<AudioProducerHttp>(_mutex, _buffer, _sampleRate);
+  ap->queueHttpGetAudio(url, gain, loop);
+  {
+    std::lock_guard<std::mutex> lock(_mutex);
+    _apQueue.push(ap);
+    startIt = _apQueue.size() == 1;
+  }
+
+  if (startIt) {
+    try {
+      ap->start(std::bind(&Track::onPlayDone, this, std::placeholders::_1, std::placeholders::_2));
+    } catch (std::exception& e) {
+      onPlayDone(true, e.what());
+    }
+  }
+}
+
+void Track::queueHttpPostAudio(const std::string& url, int gain, bool loop) {
+  bool startIt = false;
+  if (_stopping) return;
+  auto ap = std::make_shared<AudioProducerHttp>(_mutex, _buffer, _sampleRate);
+  ap->queueHttpPostAudio(url, gain, loop);
+  {
+    std::lock_guard<std::mutex> lock(_mutex);
+    _apQueue.push(ap);
+    startIt = _apQueue.size() == 1;
+  }
+
+  if (startIt) {
+    try {
+      ap->start(std::bind(&Track::onPlayDone, this, std::placeholders::_1, std::placeholders::_2));
+    } catch (std::exception& e) {
+      onPlayDone(true, e.what());
+    }
+  }
+}
+
+void Track::queueHttpPostAudio(const std::string& url, const std::string& body, std::vector<std::string>& headers, const std::string& proxy, int gain, bool loop) {
+  bool startIt = false;
+  if (_stopping) return;
+  auto ap = std::make_shared<AudioProducerHttp>(_mutex, _buffer, _sampleRate);
+    ap->queueHttpPostAudio(url, body, headers, proxy, gain, loop);
+  {
+    std::lock_guard<std::mutex> lock(_mutex);
+    _apQueue.push(ap);
+    startIt = _apQueue.size() == 1;
+  }
+
+  if (startIt) {
+    try {
+      ap->start(std::bind(&Track::onPlayDone, this, std::placeholders::_1, std::placeholders::_2));
+    } catch (std::exception& e) {
+      onPlayDone(true, e.what());
+    }
+  }
+}
+
+ bool Track::hasAudio() {
+  if (_stopping) return false;
+  std::lock_guard<std::mutex> lock(_mutex); 
+  return hasAudio_NoLock();
+}
+
+void Track::removeAllAudio() {
+  _stopping = true;
+  std::queue<std::shared_ptr<AudioProducer>> apQueueCopy;
+  {
+    std::lock_guard<std::mutex> lock(_mutex); 
+    apQueueCopy = _apQueue;
+    _apQueue = std::queue<std::shared_ptr<AudioProducer>>();
+  }
+  
+  while (!apQueueCopy.empty()) {
+    auto ap = apQueueCopy.front();
+    apQueueCopy.pop();
+    ap->stop();
+  }
+}

mod_dub/track.h

@@ -0,0 +1,52 @@
+#ifndef  __TRACK_H__
+#define __TRACK_H__
+
+#include <mutex>
+#include <queue>
+#include "common.h"
+#include "ap.h"
+
+class Track {
+public:
+  Track(const std::string& trackName, int sampleRate);
+  ~Track();
+
+  /* audio production methods */
+  void queueHttpGetAudio(const std::string& url, int gain = 0, bool loop = false);
+  void queueHttpPostAudio(const std::string& url, int gain = 0, bool loop = false);
+  void queueHttpPostAudio(const std::string& url, const std::string& body, std::vector<std::string>& headers, const std::string& proxy, int gain = 0, bool loop = false);
+  void queueFileAudio(const std::string& path, int gain = 0, bool loop = false);
+  void removeAllAudio();
+
+  void onPlayDone(bool hasError, const std::string& errMsg);
+
+  std::string& getTrackName() { return _trackName; }
+
+  /* audio playout methods */
+  bool hasAudio();
+  inline bool hasAudio_NoLock() const {
+    return !_stopping && !_buffer.empty();
+  }
+
+
+  int retrieveAndClearAudio(int16_t* buf, int desiredSamples) {
+    std::lock_guard<std::mutex> lock(_mutex);
+    int samplesToCopy = std::min(static_cast<int>(_buffer.size()), desiredSamples);
+    std::copy_n(_buffer.begin(), samplesToCopy, buf);
+    _buffer.erase(_buffer.begin(), _buffer.begin() + samplesToCopy);
+    return samplesToCopy;
+  }
+
+private:
+  std::string _trackName;
+  int _sampleRate;
+  std::mutex _mutex;
+  CircularBuffer_t _buffer;
+  std::queue<std::shared_ptr<AudioProducer>> _apQueue;
+  bool _stopping;
+};
+
+
+
+
+#endif

mod_dub/tts_vendor_parser.cpp

@@ -0,0 +1,542 @@
+#include "tts_vendor_parser.h"
+
+#include <sstream>
+#include <switch_json.h>
+#include <map>
+
+
+switch_status_t custom_vendor_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers) {
+  std::string auth_token;
+  std::string voice_name;
+  std::string custom_tts_url;
+  std::string language;
+
+  for (const auto& pair : params) {
+    if (pair.first == "auth_token") {
+      auth_token = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    } else if (pair.first == "custom_tts_url") {
+      custom_tts_url = pair.second;
+    } else if (pair.first == "language") {
+      language = pair.second;
+    }
+  }
+
+  if (custom_tts_url.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "custom_vendor_parse_text: no custom_tts_url provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+
+  url = custom_tts_url;
+
+  /* create the JSON body */
+  cJSON * jResult = cJSON_CreateObject();
+  cJSON_AddStringToObject(jResult, "text", text.c_str());
+  cJSON_AddStringToObject(jResult, "type", text.substr(0, 6) == "<speak" ? "ssml" : "text");
+  cJSON_AddNumberToObject(jResult, "samplingRate", 8000);
+  if (!voice_name.empty()) {
+    cJSON_AddStringToObject(jResult, "voice", voice_name.c_str());
+  }
+  if (!language.empty()) {
+    cJSON_AddStringToObject(jResult, "language", language.c_str());
+  }
+  char* _body = cJSON_PrintUnformatted(jResult);
+  body = _body;
+
+  cJSON_Delete(jResult);
+  free(_body);
+
+  // Create headers
+  if (!auth_token.empty()) {
+    headers.push_back("Authorization: Bearer " + auth_token);
+  }
+  headers.push_back("Accept: audio/mp3");
+  headers.push_back("Content-Type: application/json");
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t rimelabs_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers) {
+  std::string api_key;
+  std::string voice_name;
+  std::string model_id;
+  std::string speed_alpha;
+  std::string reduce_latency;
+
+  for (const auto& pair : params) {
+    if (pair.first == "api_key") {
+      api_key = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    } else if (pair.first == "model_id") {
+      model_id = pair.second;
+    } else if (pair.first == "speed_alpha") {
+      speed_alpha = pair.second;
+    }  else if (pair.first == "reduce_latency") {
+      reduce_latency = pair.second;
+    }
+  }
+
+  if (api_key.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "rimelabs_parse_text: no api_key provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (voice_name.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "rimelabs_parse_text: no voice_name provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (model_id.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "rimelabs_parse_text: no model_id provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+
+  url = "https://users.rime.ai/v1/rime-tts";
+
+  /* create the JSON body */
+  cJSON * jResult = cJSON_CreateObject();
+  cJSON_AddStringToObject(jResult, "text", text.c_str());
+  cJSON_AddNumberToObject(jResult, "samplingRate", 8000);
+  if (!voice_name.empty()) {
+    cJSON_AddStringToObject(jResult, "speaker", voice_name.c_str());
+  }
+  if (!model_id.empty()) {
+    cJSON_AddStringToObject(jResult, "modelId", model_id.c_str());
+  }
+  if (!speed_alpha.empty()) {
+    cJSON_AddNumberToObject(jResult, "speedAlpha", std::strtof(speed_alpha.c_str(), nullptr));
+  }
+  if (!reduce_latency.empty()) {
+    cJSON_AddBoolToObject(jResult, "reduceLatency", !strcmp(reduce_latency.c_str(), "true") ? 1 : 0);
+  }
+  char* _body = cJSON_PrintUnformatted(jResult);
+  body = _body;
+
+  cJSON_Delete(jResult);
+  free(_body);
+
+  // Create headers
+  headers.push_back("Authorization: Bearer " + api_key);
+  headers.push_back("Accept: audio/mp3");
+  headers.push_back("Content-Type: application/json");
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t whisper_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers) {
+  std::string api_key;
+  std::string voice_name;
+  std::string model_id;
+  std::string speed;
+
+  for (const auto& pair : params) {
+    if (pair.first == "api_key") {
+      api_key = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    } else if (pair.first == "model_id") {
+      model_id = pair.second;
+    } else if (pair.first == "speed") {
+      speed = pair.second;
+    }
+  }
+
+  if (api_key.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "whisper_parse_text: no api_key provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (model_id.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "whisper_parse_text: no model_id provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+
+  url = "https://api.openai.com/v1/audio/speech";
+
+  /* create the JSON body */
+  cJSON * jResult = cJSON_CreateObject();
+  cJSON_AddStringToObject(jResult, "model", model_id.c_str());
+  cJSON_AddStringToObject(jResult, "input", text.c_str());
+  cJSON_AddStringToObject(jResult, "voice", voice_name.c_str());
+  cJSON_AddStringToObject(jResult, "response_format", "mp3");
+  if (!speed.empty()) {
+    cJSON_AddStringToObject(jResult, "speed", speed.c_str());
+  }
+  char* _body = cJSON_PrintUnformatted(jResult);
+  body = _body;
+
+  cJSON_Delete(jResult);
+  free(_body);
+
+  // Create headers
+  headers.push_back("Authorization: Bearer " + api_key);
+  headers.push_back("Content-Type: application/json");
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t azure_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers, std::string& proxy) {
+
+  std::string api_key;
+  std::string voice_name;
+  std::string language;
+  std::string region;
+  std::string endpoint;
+  std::string endpointId;
+  std::string http_proxy_ip;
+  std::string http_proxy_port;
+
+  for (const auto& pair : params) {
+    if (pair.first == "api_key") {
+      api_key = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    } else if (pair.first == "language") {
+      language = pair.second;
+    } else if (pair.first == "region") {
+      region = pair.second;
+    } else if (pair.first == "endpoint") {
+      endpoint = pair.second;
+    } else if (pair.first == "endpointId") {
+      endpointId = pair.second;
+    } else if (pair.first == "http_proxy_ip") {
+      http_proxy_ip = pair.second;
+    } else if (pair.first == "http_proxy_port") {
+      http_proxy_port = pair.second;
+    }
+  }
+
+  if (language.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "azure_parse_text: no language provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (voice_name.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "azure_parse_text: no voice_name provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+
+  if (region.empty()) {
+    region = "westus";
+  }
+    /* format url*/
+  url = !endpoint.empty() ? endpoint : "https://" + region + ".tts.speech.microsoft.com/cognitiveservices/v1";
+
+  // Body  
+  if (strncmp(text.c_str(), "<speak", 6) == 0) {
+    body = text;
+  } else {
+    std::ostringstream body_stream;
+    body_stream << "<speak version=\"1.0\" xmlns=\"http://www.w3.org/2001/10/synthesis\" xmlns:mstts=\"https://www.w3.org/2001/mstts\" xml:lang=\"" << language << "\">";
+    body_stream << "<voice name=\"" << voice_name << "\">";
+    body_stream << text;
+    body_stream << "</voice>";
+    body_stream << "</speak>";
+    body = body_stream.str();
+  }
+
+  // Create headers
+  if (!api_key.empty()) {
+    headers.push_back("Ocp-Apim-Subscription-Key: " + api_key);
+  }
+  if (!endpointId.empty()) {
+    headers.push_back("X-Microsoft-EndpointId: " + endpointId);
+  }
+  headers.push_back("Content-Type: application/ssml+xml");
+  headers.push_back("X-Microsoft-OutputFormat: audio-16khz-32kbitrate-mono-mp3");
+
+  // Proxy
+  std::ostringstream proxy_stream;
+  if (!http_proxy_ip.empty()) {
+    proxy_stream << "http://" << http_proxy_ip;
+    if (!http_proxy_port.empty()) {
+      proxy_stream << ":" << http_proxy_port;
+    }
+  }
+  proxy = proxy_stream.str();
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t deepgram_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers) {
+
+  std::string api_key;
+  std::string voice_name;
+
+  for (const auto& pair : params) {
+    if (pair.first == "api_key") {
+      api_key = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    }
+  }
+
+  if (api_key.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "deepgram_parse_text: no api_key provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (voice_name.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "deepgram_parse_text: no voice_name provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+
+  /* format url*/
+  std::ostringstream url_stream;
+  url_stream << "https://api.deepgram.com/v1/speak?model=" << voice_name << "&encoding=mp3";
+  url = url_stream.str();
+
+  /* create the JSON body */
+  cJSON * jResult = cJSON_CreateObject();
+  cJSON_AddStringToObject(jResult, "text", text.c_str());
+
+  char* _body = cJSON_PrintUnformatted(jResult);
+  body = _body;
+
+  cJSON_Delete(jResult);
+  free(_body);
+
+  // Create headers
+  headers.push_back("Authorization: Token " + api_key);
+  headers.push_back("Content-Type: application/json");
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t playht_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers) {
+
+  std::string api_key;
+  std::string voice_name;
+  std::string user_id;
+  std::string quality;
+  std::string speed;
+  std::string seed;
+  std::string temperature;
+  std::string voice_engine;
+  std::string emotion;
+  std::string voice_guidance;
+  std::string style_guidance;
+  std::string text_guidance;
+
+  for (const auto& pair : params) {
+    if (pair.first == "api_key") {
+      api_key = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    } else if (pair.first == "user_id") {
+      user_id = pair.second;
+    } else if (pair.first == "quality") {
+      quality = pair.second;
+    } else if (pair.first == "speed") {
+      speed = pair.second;
+    } else if (pair.first == "seed") {
+      seed = pair.second;
+    } else if (pair.first == "temperature") {
+      temperature = pair.second;
+    } else if (pair.first == "voice_engine") {
+      voice_engine = pair.second;
+    }  else if (pair.first == "emotion") {
+      emotion = pair.second;
+    }  else if (pair.first == "voice_guidance") {
+      voice_guidance = pair.second;
+    }  else if (pair.first == "style_guidance") {
+      style_guidance = pair.second;
+    }  else if (pair.first == "text_guidance") {
+      text_guidance = pair.second;
+    }
+  }
+
+  if (api_key.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "playht_parse_text: no api_key provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (user_id.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "playht_parse_text: no user_id provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (voice_name.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "playht_parse_text: no voice_name provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+
+  // URL
+  url = "https://api.play.ht/api/v2/tts/stream";
+
+  /* create the JSON body */
+  cJSON * jResult = cJSON_CreateObject();
+  cJSON_AddStringToObject(jResult, "text", text.c_str());
+  cJSON_AddStringToObject(jResult, "voice", voice_name.c_str());
+  cJSON_AddStringToObject(jResult, "output_format", "mp3");
+  cJSON_AddNumberToObject(jResult, "sample_rate", 8000);
+  if (!voice_engine.empty()) {
+    cJSON_AddStringToObject(jResult, "voice_engine", voice_engine.c_str());
+  }
+  if (!quality.empty()) {
+    cJSON_AddStringToObject(jResult, "quality", quality.c_str());
+  }
+  if (!speed.empty()) {
+      double val = strtod(speed.c_str(), NULL);
+      if (val != 0.0) {
+        cJSON_AddNumberToObject(jResult, "speed", val);
+      }
+  }
+  if (!seed.empty()) {
+    cJSON_AddNumberToObject(jResult, "seed", atoi(seed.c_str()));
+  }
+  if (!temperature.empty()) {
+    cJSON_AddNumberToObject(jResult, "temperature", std::strtof(temperature.c_str(), nullptr));
+  }
+  if (!emotion.empty()) {
+    cJSON_AddStringToObject(jResult, "emotion", emotion.c_str());
+  }
+  if (!voice_guidance.empty()) {
+    cJSON_AddNumberToObject(jResult, "voice_guidance", atoi(voice_guidance.c_str()));
+  }
+  if (!style_guidance.empty()) {
+    cJSON_AddNumberToObject(jResult, "style_guidance", atoi(style_guidance.c_str()));
+  }
+  if (!text_guidance.empty()) {
+    cJSON_AddNumberToObject(jResult, "text_guidance", atoi(text_guidance.c_str()));
+  }
+  char* _body = cJSON_PrintUnformatted(jResult);
+  body = _body;
+
+  cJSON_Delete(jResult);
+  free(_body);
+
+  // Create headers
+  headers.push_back("AUTHORIZATION: " + api_key);
+  headers.push_back("X-USER-ID: " + user_id);
+  headers.push_back("Accept: audio/mpeg");
+  headers.push_back("Content-Type: application/json");
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t elevenlabs_parse_text(const std::map<std::string, std::string>& params, const std::string& text, 
+  std::string& url, std::string& body, std::vector<std::string>& headers) {
+
+  std::string api_key;
+  std::string voice_name;
+  std::string model_id;
+  std::string similarity_boost;
+  std::string stability;
+  std::string style;
+  std::string use_speaker_boost;
+  std::string optimize_streaming_latency;
+
+  for (const auto& pair : params) {
+    if (pair.first == "api_key") {
+      api_key = pair.second;
+    } else if (pair.first == "voice") {
+      voice_name = pair.second;
+    } else if (pair.first == "model_id") {
+      model_id = pair.second;
+    } else if (pair.first == "similarity_boost") {
+      similarity_boost = pair.second;
+    } else if (pair.first == "stability") {
+      stability = pair.second;
+    } else if (pair.first == "style") {
+      style = pair.second;
+    } else if (pair.first == "use_speaker_boost") {
+      use_speaker_boost = pair.second;
+    } else if (pair.first == "modeloptimize_streaming_latency_id") {
+      optimize_streaming_latency = pair.second;
+    }
+  }
+
+  if (api_key.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "elevenlabs_parse_text: no api_key provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (model_id.empty()) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "elevenlabs_parse_text: no model_id provided\n");
+    return SWITCH_STATUS_FALSE;
+  }
+  if (optimize_streaming_latency.empty()) {
+    optimize_streaming_latency = "2";
+  }
+
+  // URL
+  std::ostringstream url_stream;
+  url_stream << "https://api.elevenlabs.io/v1/text-to-speech/" << voice_name << "/stream?";
+  url_stream << "optimize_streaming_latency=" << optimize_streaming_latency << "&output_format=mp3_44100_128";
+  url = url_stream.str();
+
+  /* create the JSON body */
+  cJSON * jResult = cJSON_CreateObject();
+  cJSON_AddStringToObject(jResult, "model_id", model_id.c_str());
+  cJSON_AddStringToObject(jResult, "text", text.c_str());
+  if (!similarity_boost.empty() || !style.empty() || !use_speaker_boost.empty() || !stability.empty()) {
+    cJSON * jVoiceSettings = cJSON_CreateObject();
+    cJSON_AddItemToObject(jResult, "voice_settings", jVoiceSettings);
+    if (!similarity_boost.empty()) {
+      cJSON_AddStringToObject(jVoiceSettings, "similarity_boost", similarity_boost.c_str());
+    }
+    if (!style.empty()) {
+      cJSON_AddStringToObject(jVoiceSettings, "style", style.c_str());
+    }
+    if (!use_speaker_boost.empty()) {
+      cJSON_AddStringToObject(jVoiceSettings, "use_speaker_boost", use_speaker_boost.c_str());
+    }
+    if (!stability.empty()) {
+      cJSON_AddStringToObject(jVoiceSettings, "stability", stability.c_str());
+    }
+  }
+  char* _body = cJSON_PrintUnformatted(jResult);
+  body = _body;
+
+  cJSON_Delete(jResult);
+  free(_body);
+
+  // Create headers
+  headers.push_back("xi-api-key: " + api_key);
+  headers.push_back("Content-Type: application/json");
+
+  return SWITCH_STATUS_SUCCESS;
+}
+
+switch_status_t tts_vendor_parse_text(const std::string& say, std::string& url, std::string& body, std::vector<std::string>& headers, std::string& proxy) {
+  size_t start = say.find("{") + 1;
+  size_t end = say.find("}");
+
+  std::string text = say.substr(end + 1);
+
+  std::string params_string = say.substr(start, end - start);
+  std::istringstream ss(params_string);
+  std::map<std::string, std::string> params;
+
+  while (ss.good()) {
+    std::string substr;
+    getline(ss, substr, ',');
+    substr.erase(0, substr.find_first_not_of(' '));
+
+    size_t equal_pos = substr.find("=");
+    std::string key = substr.substr(0, equal_pos);
+    std::string value = substr.substr(equal_pos + 1, substr.size());
+
+    params[key] = value;
+  }
+
+  if (params["vendor"] == "elevenlabs") {
+    return elevenlabs_parse_text(params, text, url, body, headers);
+  } else if (params["vendor"] == "deepgram") {
+    return deepgram_parse_text(params, text, url, body, headers);
+  } else if (params["vendor"] == "microsoft") {
+    return azure_parse_text(params, text, url, body, headers, proxy);
+  } else if (params["vendor"] == "whisper") {
+    return whisper_parse_text(params, text, url, body, headers);
+  } else if (params["vendor"] == "playht") {
+    return playht_parse_text(params, text, url, body, headers);
+  } else if (params["vendor"] == "rimelabs") {
+    return rimelabs_parse_text(params, text, url, body, headers);
+  } else if (params["vendor"] == "custom") {
+    return custom_vendor_parse_text(params, text, url, body, headers);
+  }  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "tts_vendor_parse_text: There is no available parser for vendor %s\n", params["vendor"]);
+    return SWITCH_STATUS_FALSE;
+  }
+}

mod_dub/tts_vendor_parser.h

@@ -0,0 +1,12 @@
+#ifndef __MOD_DUB_TTS_VENDOR_PARSER_H__
+#define __MOD_DUB_TTS_VENDOR_PARSER_H__
+
+#include <string>
+#include <vector>
+#include <switch.h>
+#include "common.h"
+
+
+switch_status_t tts_vendor_parse_text(const std::string& say, std::string& url, std::string& body, std::vector<std::string>& headers, std::string& proxy);
+
+#endif

mod_dub/vector_math.cpp

@@ -0,0 +1,214 @@
+#include "vector_math.h"
+#include <assert.h>
+#include <string.h>
+#include <cstdlib>
+
+
+#define GRANULAR_VOLUME_MAX (50)
+#define SMAX 32767
+#define SMIN (-32768)
+#define normalize_to_16bit_basic(n) if (n > SMAX) n = SMAX; else if (n < SMIN) n = SMIN;
+#define normalize_volume_granular(x) if (x > GRANULAR_VOLUME_MAX) x = GRANULAR_VOLUME_MAX; if (x < -GRANULAR_VOLUME_MAX) x = -GRANULAR_VOLUME_MAX;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#if defined(USE_AVX2)
+#include <immintrin.h>
+#pragma message("Using AVX2 SIMD.")
+void vector_add(int16_t* a, int16_t* b, size_t len) {
+    size_t i = 0;
+    for (; i + 15 < len; i += 16) {
+        __m256i va = _mm256_loadu_si256((const __m256i*)(a + i));
+        __m256i vb = _mm256_loadu_si256((const __m256i*)(b + i));
+        __m256i vc = _mm256_add_epi16(va, vb);
+        _mm256_storeu_si256((__m256i*)(a + i), vc);
+    }
+    for (; i < len; ++i) {
+        a[i] += b[i];
+    }
+}
+void vector_normalize(int16_t* a, size_t len) {
+  __m256i max_val = _mm256_set1_epi16(SMAX);
+  __m256i min_val = _mm256_set1_epi16(SMIN);
+
+  size_t i = 0;
+  for (; i + 15 < len; i += 16) {
+      __m256i values = _mm256_loadu_si256((__m256i*)(a + i));
+      __m256i gt_max = _mm256_cmpgt_epi16(values, max_val);
+      __m256i lt_min = _mm256_cmpgt_epi16(min_val, values);
+      values = _mm256_blendv_epi8(values, max_val, gt_max);
+      values = _mm256_blendv_epi8(values, min_val, lt_min);
+      _mm256_storeu_si256((__m256i*)(a + i), values);
+  }
+
+  // Process remaining elements
+  for (; i < len; ++i) {
+      if (a[i] > SMAX) a[i] = SMAX;
+      else if (a[i] < SMIN) a[i] = SMIN;
+  }
+}
+typedef union {
+    int16_t* data;
+    __m256i* fp_avx2;
+} vector_data_t;
+
+void vector_change_sln_volume_granular(int16_t* data, uint32_t samples, int32_t vol) {
+  float newrate = 0;
+  static const float pos[GRANULAR_VOLUME_MAX] = {
+		1.122018,   1.258925,   1.412538,   1.584893,   1.778279,   1.995262,   2.238721,   2.511887,   2.818383,   3.162278,
+		3.548134,   3.981072,   4.466835,   5.011872,   5.623413,   6.309574,   7.079458,   7.943282,   8.912509,  10.000000,
+		11.220183,  12.589254,  14.125375,  15.848933,  17.782795,  19.952621,  22.387213,  25.118862,  28.183832,  31.622776,
+		35.481335,  39.810719,  44.668358,  50.118729,  56.234131,  63.095726,  70.794586,  79.432816,  89.125107, 100.000000,
+		112.201836, 125.892517, 141.253784, 158.489334, 177.827942, 199.526215, 223.872070, 251.188705, 281.838318, 316.227753
+  };
+  static const float neg[GRANULAR_VOLUME_MAX] = {
+		0.891251, 0.794328, 0.707946, 0.630957, 0.562341, 0.501187, 0.446684, 0.398107, 0.354813, 0.316228,
+		0.281838, 0.251189, 0.223872, 0.199526, 0.177828, 0.158489, 0.141254, 0.125893, 0.112202, 0.100000,
+		0.089125, 0.079433, 0.070795, 0.063096, 0.056234, 0.050119, 0.044668, 0.039811, 0.035481, 0.031623,
+		0.028184, 0.025119, 0.022387, 0.019953, 0.017783, 0.015849, 0.014125, 0.012589, 0.011220, 0.010000,
+		0.008913, 0.007943, 0.007079, 0.006310, 0.005623, 0.005012, 0.004467, 0.003981, 0.003548, 0.000000  // NOTE mapped -50 dB ratio to total silence instead of 0.003162
+  };
+  const float* chart;
+  uint32_t i = abs(vol) - 1;
+
+  if (vol == 0) return;
+  normalize_volume_granular(vol);
+
+  chart = vol > 0 ? pos : neg;
+
+  newrate = chart[i];
+
+  if (newrate) {
+    __m256 scale_factor_reg = _mm256_set1_ps(newrate);
+    uint32_t processed_samples = samples - (samples % 8); // Ensure we process only multiples of 8
+    for (uint32_t i = 0; i < processed_samples; i += 8) {
+      __m128i data_ = _mm_loadu_si128((__m128i*)(data + i));
+      __m256i data_32 = _mm256_cvtepi16_epi32(data_);
+
+      __m256 data_float = _mm256_cvtepi32_ps(data_32);
+      __m256 result = _mm256_mul_ps(data_float, scale_factor_reg);
+
+      __m256i result_32 = _mm256_cvtps_epi32(result);
+
+      // Handle saturation
+      __m256i min_val = _mm256_set1_epi32(SMIN);
+      __m256i max_val = _mm256_set1_epi32(SMAX);
+      result_32 = _mm256_min_epi32(result_32, max_val);
+      result_32 = _mm256_max_epi32(result_32, min_val);
+
+      __m128i result_16 = _mm_packs_epi32(_mm256_castsi256_si128(result_32), _mm256_extractf128_si256(result_32, 1));
+
+      _mm_storeu_si128((__m128i*)(data + i), result_16);
+    }
+
+    // Process any remaining samples
+    for (uint32_t i = processed_samples; i < samples; i++) {
+      int32_t tmp = (int32_t)(data[i] * newrate);
+      tmp = tmp > SMAX ? SMAX : (tmp < SMIN ? SMIN : tmp);
+      data[i] = (int16_t)tmp;
+    }
+  }
+}
+#elif defined(USE_SSE2)
+#include <emmintrin.h>
+#pragma message("Using SSE2 SIMD.")
+void vector_add(int16_t* a, int16_t* b, size_t len) {
+  size_t i = 0;
+  for (; i + 7 < len; i += 8) {
+    __m128i va = _mm_loadu_si128((const __m128i*)(a + i));
+    __m128i vb = _mm_loadu_si128((const __m128i*)(b + i));
+    __m128i vc = _mm_add_epi16(va, vb);
+    _mm_storeu_si128((__m128i*)(a + i), vc);
+  }
+  for (; i < len; ++i) {
+    a[i] += b[i];
+  }
+}
+void vector_normalize(int16_t* a, size_t len) {
+  __m128i max_val = _mm_set1_epi16(SMAX);
+  __m128i min_val = _mm_set1_epi16(SMIN);
+
+  size_t i = 0;
+  for (; i + 7 < len; i += 8) {
+    __m128i values = _mm_loadu_si128((__m128i*)(a + i));
+    __m128i gt_max = _mm_cmpgt_epi16(values, max_val);
+    __m128i lt_min = _mm_cmpgt_epi16(min_val, values);
+    __m128i max_masked = _mm_and_si128(gt_max, max_val);
+    __m128i min_masked = _mm_and_si128(lt_min, min_val);
+    __m128i other_masked = _mm_andnot_si128(_mm_or_si128(gt_max, lt_min), values);
+    values = _mm_or_si128(_mm_or_si128(max_masked, min_masked), other_masked);
+    _mm_storeu_si128((__m128i*)(a + i), values);
+  }
+
+  // Process remaining elements
+  for (; i < len; ++i) {
+    if (a[i] > SMAX) a[i] = SMAX;
+    else if (a[i] < SMIN) a[i] = SMIN;
+  }
+}
+
+typedef union {
+    int16_t* data;
+    __m128i* fp_sse2;
+} vector_data_t;
+
+#else
+#pragma message("Building without vector math support")
+void vector_add(int16_t* a, int16_t* b, size_t len) {
+    for (size_t i = 0; i < len; i++) {
+        a[i] += b[i];
+    }
+}
+void vector_normalize(int16_t* a, size_t len) {
+  for (size_t i = 0; i < len; i++) {
+    normalize_to_16bit_basic(a[i]);
+  }
+}
+void vector_change_sln_volume_granular(int16_t* data, uint32_t samples, int32_t vol) {
+  float newrate = 0;
+  static const float pos[GRANULAR_VOLUME_MAX] = {
+		1.122018,   1.258925,   1.412538,   1.584893,   1.778279,   1.995262,   2.238721,   2.511887,   2.818383,   3.162278,
+		3.548134,   3.981072,   4.466835,   5.011872,   5.623413,   6.309574,   7.079458,   7.943282,   8.912509,  10.000000,
+		11.220183,  12.589254,  14.125375,  15.848933,  17.782795,  19.952621,  22.387213,  25.118862,  28.183832,  31.622776,
+		35.481335,  39.810719,  44.668358,  50.118729,  56.234131,  63.095726,  70.794586,  79.432816,  89.125107, 100.000000,
+		112.201836, 125.892517, 141.253784, 158.489334, 177.827942, 199.526215, 223.872070, 251.188705, 281.838318, 316.227753
+  };
+  static const float neg[GRANULAR_VOLUME_MAX] = {
+		0.891251, 0.794328, 0.707946, 0.630957, 0.562341, 0.501187, 0.446684, 0.398107, 0.354813, 0.316228,
+		0.281838, 0.251189, 0.223872, 0.199526, 0.177828, 0.158489, 0.141254, 0.125893, 0.112202, 0.100000,
+		0.089125, 0.079433, 0.070795, 0.063096, 0.056234, 0.050119, 0.044668, 0.039811, 0.035481, 0.031623,
+		0.028184, 0.025119, 0.022387, 0.019953, 0.017783, 0.015849, 0.014125, 0.012589, 0.011220, 0.010000,
+		0.008913, 0.007943, 0.007079, 0.006310, 0.005623, 0.005012, 0.004467, 0.003981, 0.003548, 0.000000  // NOTE mapped -50 dB ratio to total silence instead of 0.003162
+  };
+  const float* chart;
+  uint32_t i;
+
+  if (vol == 0) return;
+
+  normalize_volume_granular(vol);
+
+  chart = vol > 0 ? pos : neg;
+
+  i = abs(vol) - 1;
+  assert(i < GRANULAR_VOLUME_MAX);
+  newrate = chart[i];
+  if (newrate) {
+    int32_t tmp;
+    uint32_t x;
+    int16_t *fp = data;
+
+    for (x = 0; x < samples; x++) {
+      tmp = (int32_t) (fp[x] * newrate);
+      normalize_to_16bit_basic(tmp);
+      fp[x] = (int16_t) tmp;
+    }
+  }
+}
+
+#endif
+
+#ifdef __cplusplus
+}
+#endif

mod_dub/vector_math.h

@@ -0,0 +1,20 @@
+#ifndef VECTOR_MATH_H
+#define VECTOR_MATH_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void vector_add(int16_t* a, int16_t* b, size_t len);
+void vector_normalize(int16_t* a, size_t len);
+void vector_change_sln_volume_granular(int16_t* data, uint32_t samples, int32_t vol);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif

mod_elevenlabs_tts/README.md

@@ -1,11 +1,11 @@
 # mod_google_tts
 
-A Freeswitch module that allows Google Text-to-Speech API to be used as a tts provider.
+A Freeswitch module that allows Eleven Labs' Text-to-Speech API to be used as a tts provider.
 
 ## API
 
 ### Commands
-This freeswitch module does not add any new commands, per se.  Rather, it integrates into the Freeswitch TTS interface such that it is invoked when an application uses the mod_dptools `speak` command with a tts engine of `google_tts` and a voice equal to the language code associated to one of the [supported Wavenet voices](https://cloud.google.com/text-to-speech/docs/voices)
+This freeswitch module does not add any new commands, per se.  Rather, it integrates into the Freeswitch TTS interface such that it is invoked when an application uses the mod_dptools `speak` command with a tts engine of `elevenlabs` and a voice equal to the language code associated to one of the [supported Eleven Labs voices](https://elevenlabs.io/docs/api-reference/query-library)
 
 ### Events
 None.
@@ -13,11 +13,19 @@ None.
 ## Usage
 When using [drachtio-fsrmf](https://www.npmjs.com/package/drachtio-fsmrf), you can access this functionality via the speak method on the 'endpoint' object.
 ```js
-ep.speak({
-    ttsEngine: 'google_tts',
-    voice: 'en-GB-Wavenet-A',
-    text: 'This aggression will not stand'
-  });
+var text = "Hello World";
+await endpoint.speak({
+    "ttsEngine": 'elevenlabs',
+    "voice": "W9OIfHh5DtdYiZUcFiql",
+    "text": `{use_speaker_boost=1,optimize_streaming_latency=4,style=0.5,stability=0.5,similarity_boost=0.75,api_key=XXYYZZ,model_id=eleven_turbo_v2}${text}`,
+});
 ```
-## Examples
-[google_tts.js](../../examples/google_tts.js)
+## Options
+
+Documentation on these options can be found in Eleven Labs API docs: [Voice Settings](https://elevenlabs.io/docs/speech-synthesis/voice-settings)
+
+- use_speaker_boost
+- optimize_streaming_latency
+- style
+- stability
+- similarity_boost

mod_elevenlabs_tts/elevenlabs_glue.cpp

@@ -29,8 +29,10 @@
 #include <boost/asio.hpp>
 #include <boost/assign/list_of.hpp>
 #include <boost/algorithm/string.hpp>
+#include <boost/unordered_map.hpp>
 
 #include "mod_elevenlabs_tts.h"
+#include <speex/speex_resampler.h>
 
 #define TXNID_LEN (255)
 #define URL_LEN (1024)
@@ -166,6 +168,27 @@ int multi_timer_cb(CURLM *multi, long timeout_ms, GlobalInfo_t *g) {
   return 0;
 }
 
+static std::string getEnvVar(const std::string& varName) {
+    const char* val = std::getenv(varName.c_str());
+    return val ? std::string(val) : "";
+}
+
+static long getConnectionTimeout() {
+    std::string connectTimeoutStr = getEnvVar("ELEVENLABS_TTS_CURL_CONNECT_TIMEOUT");
+
+    if (connectTimeoutStr.empty()) {
+        connectTimeoutStr = getEnvVar("TTS_CURL_CONNECT_TIMEOUT");
+    }
+
+    long connectTimeout = 10L;
+
+    if (!connectTimeoutStr.empty()) {
+        connectTimeout = std::stol(connectTimeoutStr);
+    }
+
+    return connectTimeout;
+}
+
 static CURL* createEasyHandle(void) {
   CURL* easy = curl_easy_init();
   if(!easy) {
@@ -178,7 +201,7 @@ static CURL* createEasyHandle(void) {
 
   // set connect timeout to 3 seconds and total timeout to 109 seconds
   curl_easy_setopt(easy, CURLOPT_CONNECTTIMEOUT_MS, 3000L);
-  curl_easy_setopt(easy, CURLOPT_TIMEOUT, 10L);
+  curl_easy_setopt(easy, CURLOPT_TIMEOUT, getConnectionTimeout());
 
   return easy ;    
 }
@@ -443,15 +466,12 @@ static size_t write_cb(void *ptr, size_t size, size_t nmemb, ConnInfo_t *conn) {
   CircularBuffer_t *cBuffer = (CircularBuffer_t *) el->circularBuffer;
   std::vector<uint16_t> pcm_data;
   
-  if (conn->flushed) {
+  if (conn->flushed || cBuffer == nullptr) {
     /* this will abort the transfer */
     return 0;
   }
   {
     switch_mutex_lock(el->mutex);
-    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "write_cb: received data, response %ld\n", 
-      el->response_code);
-
     if (el->response_code > 0 && el->response_code != 200) {
       std::string body((char *) ptr, bytes_received);
       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: received body %s\n", body.c_str());
@@ -533,6 +553,7 @@ static size_t write_cb(void *ptr, size_t size, size_t nmemb, ConnInfo_t *conn) {
 
           switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_time_to_first_byte_ms", time_to_first_byte_ms.c_str());
           switch_event_fire(&event);
+          el->playback_start_sent = 1;
         }
         else {
           switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: failed to create event\n");
@@ -550,9 +571,28 @@ static size_t write_cb(void *ptr, size_t size, size_t nmemb, ConnInfo_t *conn) {
   return bytes_received;
 }
 
+static int extract_response_code(const std::string& input) {
+  std::size_t space_pos = input.find(' ');
+  if (space_pos == std::string::npos) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Invalid HTTP response format %s\n", input.c_str());
+    return 0;
+  }
+
+  std::size_t code_start_pos = space_pos + 1;
+  std::size_t code_end_pos = input.find(' ', code_start_pos);
+  if (code_end_pos == std::string::npos) {
+    code_end_pos = input.length();
+  }
+
+  std::string code_str = input.substr(code_start_pos, code_end_pos - code_start_pos);
+  int response_code = std::stoi(code_str);
+  return response_code;
+}
+
+
 static size_t header_callback(char *buffer, size_t size, size_t nitems, ConnInfo_t *conn) {
   size_t bytes_received = size * nitems;
-  const std::string prefix = "HTTP/2 ";
+  const std::string prefix = "HTTP/ ";
   elevenlabs_t* el = conn->elevenlabs;
   std::string header, value;
   std::string input(buffer, bytes_received);
@@ -563,11 +603,11 @@ static size_t header_callback(char *buffer, size_t size, size_t nitems, ConnInfo
     else if (0 == header.compare("history-item-id")) el->history_item_id = strdup(value.c_str());
   }
   else {
-    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "recv header: %s\n", input.c_str());
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: %s\n", input.c_str());
     if (input.rfind(prefix, 0) == 0) {
       try {
-        el->response_code = std::stoi(input.substr(prefix.length()));
-        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "parsed response code: %ld\n", el->response_code);
+        el->response_code = extract_response_code(input);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "header_callback: parsed response code: %ld\n", el->response_code);
       } catch (const std::invalid_argument& e) {
         switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "header_callback: invalid response code %s\n", input.substr(prefix.length()).c_str());
       }
@@ -660,14 +700,14 @@ extern "C" {
     /* create temp folder for cache files */
     const char* baseDir = std::getenv("JAMBONZ_TMP_CACHE_FOLDER");
     if (!baseDir) {
-      baseDir = "/var/";
+      baseDir = "/tmp/";
     }
     if (strcmp(baseDir, "/") == 0) {
       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "failed to create folder %s\n", baseDir);
       return SWITCH_STATUS_FALSE;
     }
 
-    fullDirPath = std::string(baseDir) + "jambonz-tts-cache-files";
+    fullDirPath = std::string(baseDir) + "tts-cache-files";
 
     // Create the directory with read, write, and execute permissions for everyone
     mode_t oldMask = umask(0);
@@ -790,6 +830,12 @@ extern "C" {
     cJSON * jResult = cJSON_CreateObject();
     cJSON_AddStringToObject(jResult, "model_id", el->model_id);
     cJSON_AddStringToObject(jResult, "text", text);
+    if (el->previous_text) {
+      cJSON_AddStringToObject(jResult, "previous_text", el->previous_text);
+    }
+    if (el->next_text) {
+      cJSON_AddStringToObject(jResult, "next_text", el->next_text);
+    }
     if (el->similarity_boost || el->style || el->use_speaker_boost || el->stability) {
       cJSON * jVoiceSettings = cJSON_CreateObject();
       cJSON_AddItemToObject(jResult, "voice_settings", jVoiceSettings);
@@ -830,6 +876,15 @@ extern "C" {
 
     el->circularBuffer = (void *) new CircularBuffer_t(8192);
 
+    if (el->rate != 8000 /*Hz*/) {
+      int err;
+      el->resampler = speex_resampler_init(1, 8000, el->rate, SWITCH_RESAMPLE_QUALITY, &err);
+      if (0 != err) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing resampler: %s.\n", speex_resampler_strerror(err));
+        return SWITCH_STATUS_FALSE;
+      }
+    }
+
     std::ostringstream api_key_stream;
     api_key_stream << "xi-api-key: " << el->api_key;
 
@@ -856,7 +911,10 @@ extern "C" {
     curl_easy_setopt(easy, CURLOPT_POSTFIELDS, conn->body);
     //curl_easy_setopt(easy, CURLOPT_POSTFIELDSIZE, body.length());
 
-    curl_easy_setopt(easy, CURLOPT_HTTP_VERSION, CURL_HTTP_VERSION_2_0);
+    // libcurl adding random byte to the response body that creates white noise to audio file
+    // https://github.com/curl/curl/issues/10525
+    const bool disable_http_2 = switch_true(std::getenv("DISABLE_HTTP2_FOR_TTS_STREAMING"));
+    curl_easy_setopt(easy, CURLOPT_HTTP_VERSION, disable_http_2 ? CURL_HTTP_VERSION_1_1 : CURL_HTTP_VERSION_2_0);
 
     rc = curl_multi_add_handle(global.multi, conn->easy);
     mcode_test("new_conn: curl_multi_add_handle", rc);
@@ -881,7 +939,6 @@ extern "C" {
     {
       switch_mutex_lock(el->mutex);
       ConnInfo_t *conn = (ConnInfo_t *) el->conn;
-
       if (el->response_code > 0 && el->response_code != 200) {
         switch_mutex_unlock(el->mutex);
         switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "elevenlabs_speech_read_tts, returning failure\n") ;  
@@ -901,14 +958,33 @@ extern "C" {
         switch_mutex_unlock(el->mutex);
         return SWITCH_STATUS_SUCCESS;
       }
-      size_t size = std::min((*datalen/2), cBuffer->size());
+      size_t size = std::min((*datalen/(2 * el->rate / 8000)), cBuffer->size());
       pcm_data.insert(pcm_data.end(), cBuffer->begin(), cBuffer->begin() + size);
       cBuffer->erase(cBuffer->begin(), cBuffer->begin() + size);
       switch_mutex_unlock(el->mutex);
     }
 
-    memcpy(data, pcm_data.data(), pcm_data.size() * sizeof(uint16_t));
-    *datalen = pcm_data.size() * sizeof(uint16_t);
+    size_t data_size = pcm_data.size();
+
+    if (el->resampler) {
+      std::vector<int16_t> in(pcm_data.begin(), pcm_data.end());
+
+      std::vector<int16_t> out((*datalen));
+      spx_uint32_t in_len = data_size;
+      spx_uint32_t out_len = out.size();
+      speex_resampler_process_interleaved_int(el->resampler, in.data(), &in_len, out.data(), &out_len);
+
+      if (out_len > out.size()) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, "Resampler output exceeded maximum buffer size!\n");
+        return SWITCH_STATUS_FALSE;
+      }
+
+      memcpy(data, out.data(), out_len * sizeof(int16_t));
+      *datalen = out_len * sizeof(int16_t);
+    } else {
+      memcpy(data, pcm_data.data(), pcm_data.size() * sizeof(uint16_t));
+      *datalen = pcm_data.size() * sizeof(uint16_t);
+    }
 
     return SWITCH_STATUS_SUCCESS;
   }
@@ -922,8 +998,15 @@ extern "C" {
     delete cBuffer;
     el->circularBuffer = nullptr ;
 
+    // destroy resampler
+    if (el->resampler) {
+      speex_resampler_destroy(el->resampler);
+      el->resampler = NULL;
+    }
+
     if (conn) {
       conn->flushed = true;
+      
       if (!download_complete) {
         if (conn->file) {
           switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "closing audio cache file %s because download was interrupted\n", el->cache_filename);
@@ -932,18 +1015,18 @@ extern "C" {
           }
           conn->file = nullptr ;
         }
-
-        if (el->cache_filename) {
-          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing audio cache file %s because download was interrupted\n", el->cache_filename);
-          if (unlink(el->cache_filename) != 0) {
-            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error removing audio cache file %s: %d:%s\n", 
-              el->cache_filename, errno, strerror(errno));
-          }
-          free(el->cache_filename);
-          el->cache_filename = nullptr ;
-        }
       }
     }
+    // If playback_start has not been sent, delete the file
+    if (el->cache_filename && !el->playback_start_sent) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing audio cache file %s because download was interrupted\n", el->cache_filename);
+      if (unlink(el->cache_filename) != 0) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error removing audio cache file %s: %d:%s\n", 
+          el->cache_filename, errno, strerror(errno));
+      }
+      free(el->cache_filename);
+      el->cache_filename = nullptr ;
+    }
     if (el->session_id) {
       switch_core_session_t* session = switch_core_session_locate(el->session_id);
       if (session) {

mod_elevenlabs_tts/mod_elevenlabs_tts.c

@@ -19,6 +19,8 @@ static void clearElevenlabs(elevenlabs_t* el, int freeAll) {
   if (el->style) free(el->style);
   if (el->use_speaker_boost) free(el->use_speaker_boost);
   if (el->optimize_streaming_latency) free(el->optimize_streaming_latency);
+  if (el->previous_text) free(el->previous_text);
+  if (el->next_text) free(el->next_text);
   if (el->ct) free(el->ct);
   if (el->reported_latency) free(el->reported_latency);
   if (el->request_id) free(el->request_id);
@@ -36,6 +38,8 @@ static void clearElevenlabs(elevenlabs_t* el, int freeAll) {
   el->style = NULL;
   el->use_speaker_boost = NULL;
   el->optimize_streaming_latency = NULL;
+  el->previous_text = NULL;
+  el->next_text = NULL;
   el->ct = NULL;
   el->reported_latency = NULL;
   el->request_id = NULL;
@@ -106,6 +110,9 @@ static switch_status_t ell_speech_feed_tts(switch_speech_handle_t *sh, char *tex
   elevenlabs_t *el = createOrRetrievePrivateData(sh);
   el->draining = 0;
   el->reads = 0;
+  el->response_code = 0;
+  el->err_msg = NULL;
+  el->playback_start_sent = 0;
   
   switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "ell_speech_feed_tts\n");
   
@@ -171,6 +178,14 @@ static void ell_text_param_tts(switch_speech_handle_t *sh, char *param, const ch
     if (el->optimize_streaming_latency) free(el->optimize_streaming_latency);
     el->optimize_streaming_latency = strdup(val);
   }
+  else if (0 == strcmp(param, "next_text")) {
+    if (el->next_text) free(el->next_text);
+    el->next_text = strdup(val);
+  }
+  else if (0 == strcmp(param, "previous_text")) {
+    if (el->previous_text) free(el->previous_text);
+    el->previous_text = strdup(val);
+  }
   else if (0 == strcmp(param, "write_cache_file") && switch_true(val)) {
     el->cache_audio = 1;
   }

mod_elevenlabs_tts/mod_elevenlabs_tts.h

@@ -5,6 +5,7 @@
 #include <fcntl.h>
 #include <sys/stat.h>
 #include <unistd.h>
+#include <speex/speex_resampler.h>
 
 struct elevenlabs_data {
   char *session_id;
@@ -16,6 +17,8 @@ struct elevenlabs_data {
   char* style;
   char* use_speaker_boost;
   char* optimize_streaming_latency;
+  char* previous_text;
+  char* next_text;
 
   /* result data */
   long response_code;
@@ -38,6 +41,8 @@ struct elevenlabs_data {
   int draining;
   int reads;
   int cache_audio;
+  int playback_start_sent;
+  SpeexResamplerState *resampler;
 };
 
 typedef struct elevenlabs_data elevenlabs_t;

mod_google_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_google_transcribe/Makefile.am

@@ -2,7 +2,7 @@ include $(top_srcdir)/build/modmake.rulesam
 MODNAME=mod_google_transcribe
 
 mod_LTLIBRARIES = mod_google_transcribe.la
-mod_google_transcribe_la_SOURCES  = mod_google_transcribe.c google_glue.cpp
+mod_google_transcribe_la_SOURCES  = mod_google_transcribe.c google_glue.cpp google_glue_v1.cpp google_glue_v2.cpp
 mod_google_transcribe_la_CFLAGS   = $(AM_CFLAGS)
 mod_google_transcribe_la_CXXFLAGS = -I $(top_srcdir)/libs/googleapis/gens $(AM_CXXFLAGS) -std=c++17
 

mod_google_transcribe/generic_google_glue.h

@@ -0,0 +1,248 @@
+#ifndef __GENERIC_GOOGLE_GLUE_H__
+#define __GENERIC_GOOGLE_GLUE_H__
+
+#include <switch_json.h>
+
+template<typename Streamer>
+switch_bool_t google_speech_frame(switch_media_bug_t *bug, void* user_data) {
+	switch_core_session_t *session = switch_core_media_bug_get_session(bug);
+	struct cap_cb *cb = (struct cap_cb *) user_data;
+    if (cb->streamer && (!cb->wants_single_utterance || !cb->got_end_of_utterance)) {
+        Streamer* streamer = (Streamer *) cb->streamer;
+        uint8_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
+        switch_frame_t frame = {};
+        frame.data = data;
+        frame.buflen = SWITCH_RECOMMENDED_BUFFER_SIZE;
+
+        if (switch_mutex_trylock(cb->mutex) == SWITCH_STATUS_SUCCESS) {
+            while (switch_core_media_bug_read(bug, &frame, SWITCH_TRUE) == SWITCH_STATUS_SUCCESS && !switch_test_flag((&frame), SFF_CNG)) {
+                if (frame.datalen) {
+                    if (cb->vad && !streamer->isConnected()) {
+                        switch_vad_state_t state = switch_vad_process(cb->vad, (int16_t*) frame.data, frame.samples);
+                        if (state == SWITCH_VAD_STATE_START_TALKING) {
+                            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "detected speech, connect to google speech now\n");
+                            streamer->connect();
+                            cb->responseHandler(session, "vad_detected", cb->bugname);
+                        }
+                    }
+
+                    if (cb->resampler) {
+                        spx_int16_t out[SWITCH_RECOMMENDED_BUFFER_SIZE];
+                        spx_uint32_t out_len = SWITCH_RECOMMENDED_BUFFER_SIZE;
+                        spx_uint32_t in_len = frame.samples;
+                        size_t written;
+
+                        speex_resampler_process_interleaved_int(cb->resampler,
+                            (const spx_int16_t *) frame.data,
+                            (spx_uint32_t *) &in_len,
+                            &out[0],
+                            &out_len);
+                        streamer->write( &out[0], sizeof(spx_int16_t) * out_len);
+                    }
+                    else {
+                        streamer->write( frame.data, sizeof(spx_int16_t) * frame.samples);
+                    }
+                }
+            }
+            switch_mutex_unlock(cb->mutex);
+        }
+	}
+	return SWITCH_TRUE;
+}
+
+template<typename Streamer>
+switch_status_t google_speech_session_init(switch_core_session_t *session, responseHandler_t responseHandler,
+		switch_thread_start_t func, uint32_t to_rate, uint32_t samples_per_second, uint32_t channels, char* lang,
+		int interim, char *bugname, int single_utterance, int separate_recognition, int max_alternatives,
+		int profanity_filter, int word_time_offset, int punctuation, const char* model, int enhanced,
+		const char* hints, char* play_file, void **ppUserData) {
+
+	switch_channel_t *channel = switch_core_session_get_channel(session);
+	auto read_codec = switch_core_session_get_read_codec(session);
+	uint32_t sampleRate = read_codec->implementation->actual_samples_per_second;
+	struct cap_cb *cb;
+	int err;
+
+	cb =(struct cap_cb *) switch_core_session_alloc(session, sizeof(*cb));
+	strncpy(cb->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
+	strncpy(cb->bugname, bugname, MAX_BUG_LEN);
+	cb->got_end_of_utterance = 0;
+	cb->wants_single_utterance = single_utterance;
+	if (play_file != NULL){
+		cb->play_file = 1;
+	}
+	
+	switch_mutex_init(&cb->mutex, SWITCH_MUTEX_NESTED, switch_core_session_get_pool(session));
+	if (sampleRate != to_rate) {
+		cb->resampler = speex_resampler_init(channels, sampleRate, to_rate, SWITCH_RESAMPLE_QUALITY, &err);
+	if (0 != err) {
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing resampler: %s.\n",
+								switch_channel_get_name(channel), speex_resampler_strerror(err));
+		return SWITCH_STATUS_FALSE;
+	}
+	} else {
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "%s: no resampling needed for this call\n", switch_channel_get_name(channel));
+	}
+	cb->responseHandler = responseHandler;
+
+	// allocate vad if we are delaying connecting to the recognizer until we detect speech
+	if (switch_channel_var_true(channel, "START_RECOGNIZING_ON_VAD")) {
+		cb->vad = switch_vad_init(sampleRate, channels);
+		if (cb->vad) {
+			const char* var;
+			int mode = 2;
+			int silence_ms = 150;
+			int voice_ms = 250;
+			int debug = 0;
+
+			if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_MODE")) {
+				mode = atoi(var);
+			}
+			if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_SILENCE_MS")) {
+				silence_ms = atoi(var);
+			}
+			if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_VOICE_MS")) {
+				voice_ms = atoi(var);
+			}
+			if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_VOICE_MS")) {
+				voice_ms = atoi(var);
+			}
+			switch_vad_set_mode(cb->vad, mode);
+			switch_vad_set_param(cb->vad, "silence_ms", silence_ms);
+			switch_vad_set_param(cb->vad, "voice_ms", voice_ms);
+			switch_vad_set_param(cb->vad, "debug", debug);
+		}
+	}
+
+	Streamer *streamer = NULL;
+	try {
+	    streamer = new Streamer(session, channels, lang, interim, to_rate, sampleRate, single_utterance, separate_recognition, max_alternatives,
+		    profanity_filter, word_time_offset, punctuation, model, enhanced, hints);
+	    cb->streamer = streamer;
+	} catch (std::exception& e) {
+	    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing gstreamer: %s.\n", 
+		switch_channel_get_name(channel), e.what());
+	    return SWITCH_STATUS_FALSE;
+	}
+
+	if (!cb->vad) streamer->connect();
+
+	// create the read thread
+	switch_threadattr_t *thd_attr = NULL;
+	switch_memory_pool_t *pool = switch_core_session_get_pool(session);
+
+	switch_threadattr_create(&thd_attr, pool);
+	switch_threadattr_stacksize_set(thd_attr, SWITCH_THREAD_STACKSIZE);
+	switch_thread_create(&cb->thread, thd_attr, func, cb, pool);
+
+	*ppUserData = cb;
+	return SWITCH_STATUS_SUCCESS;
+}
+
+template<typename Streamer>
+switch_status_t google_speech_session_cleanup(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug) {
+	switch_channel_t *channel = switch_core_session_get_channel(session);
+
+	if (bug) {
+		struct cap_cb *cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+		switch_mutex_lock(cb->mutex);
+
+		if (!switch_channel_get_private(channel, cb->bugname)) {
+			// race condition
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s Bug is not attached (race).\n", switch_channel_get_name(channel));
+			switch_mutex_unlock(cb->mutex);
+			return SWITCH_STATUS_FALSE;
+		}
+		switch_channel_set_private(channel, cb->bugname, NULL);
+
+		// stop playback if available
+		if (cb->play_file == 1){ 
+			if (switch_channel_test_flag(channel, CF_BROADCAST)) {
+				switch_channel_stop_broadcast(channel);
+			} else {
+				switch_channel_set_flag_value(channel, CF_BREAK, 1);
+			}
+		}
+
+		// close connection and get final responses
+		Streamer* streamer = (Streamer *) cb->streamer;
+
+		if (streamer) {
+			streamer->writesDone();
+
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "google_speech_session_cleanup: GStreamer (%p) waiting for read thread to complete\n", (void*)streamer);
+			switch_status_t st;
+			switch_thread_join(&st, cb->thread);
+			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "google_speech_session_cleanup:  GStreamer (%p) read thread completed\n", (void*)streamer);
+
+			delete streamer;
+			cb->streamer = NULL;
+		}
+
+		if (cb->resampler) {
+			speex_resampler_destroy(cb->resampler);
+		}
+		if (cb->vad) {
+			switch_vad_destroy(&cb->vad);
+			cb->vad = nullptr;
+		}
+		if (!channelIsClosing) {
+			switch_core_media_bug_remove(session, &bug);
+		}
+
+		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "google_speech_session_cleanup: Closed stream\n");
+
+		switch_mutex_unlock(cb->mutex);
+
+		return SWITCH_STATUS_SUCCESS;
+	}
+
+	switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s Bug is not attached.\n", switch_channel_get_name(channel));
+	return SWITCH_STATUS_FALSE;
+}
+
+template<typename PhraseSet>
+void google_speech_configure_grammar_hints(switch_core_session_t *session, switch_channel_t *channel, const char* hints, PhraseSet* phrase_set) {
+    float boost = -1;
+
+    // get boost setting for the phrase set in its entirety
+    if (switch_true(switch_channel_get_variable(channel, "GOOGLE_SPEECH_HINTS_BOOST"))) {
+        boost = (float) atof(switch_channel_get_variable(channel, "GOOGLE_SPEECH_HINTS_BOOST"));
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "boost value: %f\n", boost);
+        phrase_set->set_boost(boost);
+    }
+
+    // hints are either a simple comma-separated list of phrases, or a json array of objects
+    // containing a phrase and a boost value
+    auto *jHint = cJSON_Parse((char *) hints);
+    if (jHint) {
+        int i = 0;
+        cJSON *jPhrase = NULL;
+        cJSON_ArrayForEach(jPhrase, jHint) {
+            cJSON *jItem = cJSON_GetObjectItem(jPhrase, "phrase");
+            if (jItem) {
+                auto* phrase = phrase_set->add_phrases();
+                phrase->set_value(cJSON_GetStringValue(jItem));
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "phrase: %s\n", phrase->value().c_str());
+                if (cJSON_GetObjectItem(jPhrase, "boost")) {
+                    phrase->set_boost((float) cJSON_GetObjectItem(jPhrase, "boost")->valuedouble);
+                    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "boost value: %f\n", phrase->boost());
+                }
+                i++;
+            }
+        }
+        cJSON_Delete(jHint);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "added %d hints\n", i);
+    }
+    else {
+        char *phrases[500] = { 0 };
+        int argc = switch_separate_string((char *) hints, ',', phrases, 500);
+        for (int i = 0; i < argc; i++) {
+            auto* phrase = phrase_set->add_phrases();
+            phrase->set_value(phrases[i]);
+        }
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "added %d hints\n", argc);
+    }
+}
+
+#endif

mod_google_transcribe/google_glue.cpp

@@ -1,727 +1,27 @@
-#include <cstdlib>
-#include <algorithm>
-#include <future>
-
 #include <switch.h>
-#include <switch_json.h>
 #include <grpc++/grpc++.h>
-
-#include "google/cloud/speech/v1p1beta1/cloud_speech.grpc.pb.h"
-
-#include <switch_json.h>
+#include <unistd.h>
 
 #include "mod_google_transcribe.h"
-#include "simple_buffer.h"
-
-using google::cloud::speech::v1p1beta1::RecognitionConfig;
-using google::cloud::speech::v1p1beta1::Speech;
-using google::cloud::speech::v1p1beta1::SpeechContext;
-using google::cloud::speech::v1p1beta1::StreamingRecognizeRequest;
-using google::cloud::speech::v1p1beta1::StreamingRecognizeResponse;
-using google::cloud::speech::v1p1beta1::SpeakerDiarizationConfig;
-using google::cloud::speech::v1p1beta1::SpeechAdaptation;
-using google::cloud::speech::v1p1beta1::PhraseSet;
-using google::cloud::speech::v1p1beta1::PhraseSet_Phrase;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_DISCUSSION;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_PRESENTATION;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_PHONE_CALL;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_VOICEMAIL;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_PROFESSIONALLY_PRODUCED;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_VOICE_SEARCH;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_VOICE_COMMAND;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_DICTATION;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_MicrophoneDistance_NEARFIELD;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_MicrophoneDistance_MIDFIELD;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_MicrophoneDistance_FARFIELD;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_OriginalMediaType_AUDIO;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_OriginalMediaType_VIDEO;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_SMARTPHONE;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_PC;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_PHONE_LINE;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_VEHICLE;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_OTHER_OUTDOOR_DEVICE;
-using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_OTHER_INDOOR_DEVICE;
-using google::cloud::speech::v1p1beta1::StreamingRecognizeResponse_SpeechEventType_END_OF_SINGLE_UTTERANCE;
-using google::rpc::Status;
-
-#define CHUNKSIZE (320)
-
-namespace {
-  int case_insensitive_match(std::string s1, std::string s2) {
-   std::transform(s1.begin(), s1.end(), s1.begin(), ::tolower);
-   std::transform(s2.begin(), s2.end(), s2.begin(), ::tolower);
-   if(s1.compare(s2) == 0)
-      return 1; //The strings are same
-   return 0; //not matched
-  }
-}
-class GStreamer;
-
-class GStreamer {
-public:
-	GStreamer(
-    switch_core_session_t *session, 
-    uint32_t channels, 
-    char* lang, 
-    int interim, 
-    uint32_t config_sample_rate,
-		uint32_t samples_per_second,
-    int single_utterance, 
-    int separate_recognition,
-		int max_alternatives, 
-    int profanity_filter, 
-    int word_time_offset, 
-    int punctuation, 
-    const char* model, 
-    int enhanced, 
-		const char* hints) : m_session(session), m_writesDone(false), m_connected(false), 
-      m_audioBuffer(CHUNKSIZE, 15) {
-  
-    const char* var;
-    const char* google_uri;
-    switch_channel_t *channel = switch_core_session_get_channel(session);
-
-    if (!(google_uri = switch_channel_get_variable(channel, "GOOGLE_SPEECH_TO_TEXT_URI"))) {
-      google_uri = "speech.googleapis.com";
-    }
-		if (var = switch_channel_get_variable(channel, "GOOGLE_APPLICATION_CREDENTIALS")) {
-			auto channelCreds = grpc::SslCredentials(grpc::SslCredentialsOptions());
-			auto callCreds = grpc::ServiceAccountJWTAccessCredentials(var);
-			auto creds = grpc::CompositeChannelCredentials(channelCreds, callCreds);
-			m_channel = grpc::CreateChannel(google_uri, creds);
-		}
-		else {
-			auto creds = grpc::GoogleDefaultCredentials();
-			m_channel = grpc::CreateChannel(google_uri, creds);
-		}
-
-  	m_stub = Speech::NewStub(m_channel);
-  		
-		auto* streaming_config = m_request.mutable_streaming_config();
-		RecognitionConfig* config = streaming_config->mutable_config();
-
-    streaming_config->set_interim_results(interim);
-    if (single_utterance == 1) {
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_single_utterance\n");
-      streaming_config->set_single_utterance(true);
-    }
-    else {
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_single_utterance is FALSE\n");
-      streaming_config->set_single_utterance(false);
-    }
-
-		config->set_language_code(lang);
-    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "transcribe language %s \n", lang);
-    
-  	config->set_sample_rate_hertz(config_sample_rate);
-
-		config->set_encoding(RecognitionConfig::LINEAR16);
-
-    // the rest of config comes from channel vars
-
-    // number of channels in the audio stream (default: 1)
-    if (channels > 1) {
-      config->set_audio_channel_count(channels);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "audio_channel_count %d\n", channels);
-
-      // transcribe each separately?
-      if (separate_recognition == 1) {
-        config->set_enable_separate_recognition_per_channel(true);
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_separate_recognition_per_channel on\n");
-      }
-    }
-
-    // max alternatives
-    if (max_alternatives > 1) {
-      config->set_max_alternatives(max_alternatives);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "max_alternatives %d\n", max_alternatives);
-    }
-
-    // profanity filter
-    if (profanity_filter == 1) {
-      config->set_profanity_filter(true);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "profanity_filter\n");
-    }
-
-    // enable word offsets
-    if (word_time_offset == 1) {
-      config->set_enable_word_time_offsets(true);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_word_time_offsets\n");
-    }
-
-    // enable automatic punctuation
-    if (punctuation == 1) {
-      config->set_enable_automatic_punctuation(true);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_automatic_punctuation\n");
-    }
-    else {
-      config->set_enable_automatic_punctuation(false);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "disable_automatic_punctuation\n");
-    }
-
-    // speech model
-    if (model != NULL) {
-      config->set_model(model);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "speech model %s\n", model);
-    }
-
-    // use enhanced model
-    if (enhanced == 1) {
-      config->set_use_enhanced(true);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "use_enhanced\n");
-    }
-
-    // hints  
-    if (hints != NULL) {
-      auto* adaptation = config->mutable_adaptation();
-      auto* phrase_set = adaptation->add_phrase_sets();
-      auto *context = config->add_speech_contexts();
-      float boost = -1;
-
-      // get boost setting for the phrase set in its entirety
-      if (switch_true(switch_channel_get_variable(channel, "GOOGLE_SPEECH_HINTS_BOOST"))) {
-     	  boost = (float) atof(switch_channel_get_variable(channel, "GOOGLE_SPEECH_HINTS_BOOST"));
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "boost value: %f\n", boost);
-        phrase_set->set_boost(boost);
-      }
-
-      // hints are either a simple comma-separated list of phrases, or a json array of objects
-      // containing a phrase and a boost value
-      auto *jHint = cJSON_Parse((char *) hints);
-      if (jHint) {
-        int i = 0;
-        cJSON *jPhrase = NULL;
-        cJSON_ArrayForEach(jPhrase, jHint) {
-          auto* phrase = phrase_set->add_phrases();
-          cJSON *jItem = cJSON_GetObjectItem(jPhrase, "phrase");
-          if (jItem) {
-            phrase->set_value(cJSON_GetStringValue(jItem));
-            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "phrase: %s\n", phrase->value().c_str());
-            if (cJSON_GetObjectItem(jPhrase, "boost")) {
-              phrase->set_boost((float) cJSON_GetObjectItem(jPhrase, "boost")->valuedouble);
-              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "boost value: %f\n", phrase->boost());
-            }
-            i++;
-          }
-        }
-        cJSON_Delete(jHint);
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "added %d hints\n", i);
-      }
-      else {
-        char *phrases[500] = { 0 };
-        int argc = switch_separate_string((char *) hints, ',', phrases, 500);
-        for (int i = 0; i < argc; i++) {
-          auto* phrase = phrase_set->add_phrases();
-          phrase->set_value(phrases[i]);
-        }
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "added %d hints\n", argc);
-      }
-    }
-
-    // alternative language
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_ALTERNATIVE_LANGUAGE_CODES")) {
-      char *alt_langs[3] = { 0 };
-      int argc = switch_separate_string((char *) var, ',', alt_langs, 3);
-      for (int i = 0; i < argc; i++) {
-        config->add_alternative_language_codes(alt_langs[i]);
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "added alternative lang %s\n", alt_langs[i]);
-      }
-    }
-
-    // speaker diarization
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION")) {
-      auto* diarization_config = config->mutable_diarization_config();
-      diarization_config->set_enable_speaker_diarization(true);
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enabling speaker diarization\n", var);
-      if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION_MIN_SPEAKER_COUNT")) {
-        int count = std::max(atoi(var), 1);
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting min speaker count to %d\n", count);
-        diarization_config->set_min_speaker_count(count);
-      }
-      if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION_MAX_SPEAKER_COUNT")) {
-        int count = std::max(atoi(var), 2);
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting max speaker count to %d\n", count);
-        diarization_config->set_max_speaker_count(count);
-      }
-    }
-
-    // recognition metadata
-    auto* metadata = config->mutable_metadata();
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_INTERACTION_TYPE")) {
-      if (case_insensitive_match("discussion", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_DISCUSSION);
-      if (case_insensitive_match("presentation", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_PRESENTATION);
-      if (case_insensitive_match("phone_call", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_PHONE_CALL);
-      if (case_insensitive_match("voicemail", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_VOICEMAIL);
-      if (case_insensitive_match("professionally_produced", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_PROFESSIONALLY_PRODUCED);
-      if (case_insensitive_match("voice_search", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_VOICE_SEARCH);
-      if (case_insensitive_match("voice_command", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_VOICE_COMMAND);
-      if (case_insensitive_match("dictation", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_DICTATION);
-    }
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_INDUSTRY_NAICS_CODE")) {
-      metadata->set_industry_naics_code_of_audio(atoi(var));
-    }
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_MICROPHONE_DISTANCE")) {
-      if (case_insensitive_match("nearfield", var)) metadata->set_microphone_distance(RecognitionMetadata_MicrophoneDistance_NEARFIELD);
-      if (case_insensitive_match("midfield", var)) metadata->set_microphone_distance(RecognitionMetadata_MicrophoneDistance_MIDFIELD);
-      if (case_insensitive_match("farfield", var)) metadata->set_microphone_distance(RecognitionMetadata_MicrophoneDistance_FARFIELD);
-    }
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_ORIGINAL_MEDIA_TYPE")) {
-      if (case_insensitive_match("audio", var)) metadata->set_original_media_type(RecognitionMetadata_OriginalMediaType_AUDIO);
-      if (case_insensitive_match("video", var)) metadata->set_original_media_type(RecognitionMetadata_OriginalMediaType_VIDEO);
-    }
-    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_RECORDING_DEVICE_TYPE")) {
-      if (case_insensitive_match("smartphone", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_SMARTPHONE);
-      if (case_insensitive_match("pc", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_PC);
-      if (case_insensitive_match("phone_line", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_PHONE_LINE);
-      if (case_insensitive_match("vehicle", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_VEHICLE);
-      if (case_insensitive_match("other_outdoor_device", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_OTHER_OUTDOOR_DEVICE);
-      if (case_insensitive_match("other_indoor_device", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_OTHER_INDOOR_DEVICE);
-    }
-	}
-
-	~GStreamer() {
-		//switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_INFO, "GStreamer::~GStreamer - deleting channel and stub: %p\n", (void*)this);
-	}
-
-  void connect() {
-    assert(!m_connected);
-    // Begin a stream.
-  	m_streamer = m_stub->StreamingRecognize(&m_context);
-    m_connected = true;
-
-    // read thread is waiting on this
-    m_promise.set_value();
-
-  	// Write the first request, containing the config only.
-  	m_streamer->Write(m_request);
-
-    // send any buffered audio
-    int nFrames = m_audioBuffer.getNumItems();
-    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p got stream ready, %d buffered frames\n", this, nFrames);	
-    if (nFrames) {
-      char *p;
-      do {
-        p = m_audioBuffer.getNextChunk();
-        if (p) {
-          write(p, CHUNKSIZE);
-        }
-      } while (p);
-    }
-  }
-
-	bool write(void* data, uint32_t datalen) {
-    if (!m_connected) {
-      if (datalen % CHUNKSIZE == 0) {
-        m_audioBuffer.add(data, datalen);
-      }
-      return true;
-    }
-    m_request.set_audio_content(data, datalen);
-    bool ok = m_streamer->Write(m_request);
-    return ok;
-  }
-
-	uint32_t nextMessageSize(void) {
-		uint32_t size = 0;
-		m_streamer->NextMessageSize(&size);
-		return size;
-	}
-
-	bool read(StreamingRecognizeResponse* response) {
-		return m_streamer->Read(response);
-	}
-
-	grpc::Status finish() {
-		return m_streamer->Finish();
-	}
-
-	void writesDone() {
-    // grpc crashes if we call this twice on a stream
-    if (!m_connected) {
-      cancelConnect();
-    }
-    else if (!m_writesDone) {
-      m_streamer->WritesDone();
-      m_writesDone = true;
-    }
-	}
-
-  bool waitForConnect() {
-    std::shared_future<void> sf(m_promise.get_future());
-    sf.wait();
-    return m_connected;
-  }
-
-  void cancelConnect() {
-    assert(!m_connected);
-    m_promise.set_value();
-  }
-
-  bool isConnected() {
-    return m_connected;
-  }
-
-private:
-	switch_core_session_t* m_session;
-  grpc::ClientContext m_context;
-	std::shared_ptr<grpc::Channel> m_channel;
-	std::unique_ptr<Speech::Stub> 	m_stub;
-	std::unique_ptr< grpc::ClientReaderWriterInterface<StreamingRecognizeRequest, StreamingRecognizeResponse> > m_streamer;
-	StreamingRecognizeRequest m_request;
-  bool m_writesDone;
-  bool m_connected;
-  std::promise<void> m_promise;
-  SimpleBuffer m_audioBuffer;
-};
-
-static void *SWITCH_THREAD_FUNC grpc_read_thread(switch_thread_t *thread, void *obj) {
-  static int count;
-	struct cap_cb *cb = (struct cap_cb *) obj;
-	GStreamer* streamer = (GStreamer *) cb->streamer;
-
-  bool connected = streamer->waitForConnect();
-  if (!connected) {
-    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "google transcribe grpc read thread exiting since we didnt connect\n") ;
-    return nullptr;
-  }
-
-  // Read responses.
-  StreamingRecognizeResponse response;
-  while (streamer->read(&response)) {  // Returns false when no more to read.
-    switch_core_session_t* session = switch_core_session_locate(cb->sessionId);
-    if (!session) {
-      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "grpc_read_thread: session %s is gone!\n", cb->sessionId) ;
-      return nullptr;
-    }
-    count++;
-    auto speech_event_type = response.speech_event_type();
-    if (response.has_error()) {
-      Status status = response.error();
-      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "grpc_read_thread: error %s (%d)\n", status.message().c_str(), status.code()) ;
-      cJSON* json = cJSON_CreateObject();
-      cJSON_AddStringToObject(json, "type", "error");
-      cJSON_AddStringToObject(json, "error", status.message().c_str());
-      char* jsonString = cJSON_PrintUnformatted(json);
-      cb->responseHandler(session, jsonString, cb->bugname);
-      free(jsonString);
-      cJSON_Delete(json);
-    }
-    
-    if (cb->play_file == 1){
-      cb->responseHandler(session, "play_interrupt", cb->bugname);
-    }
-    
-    for (int r = 0; r < response.results_size(); ++r) {
-      auto result = response.results(r);
-      cJSON * jResult = cJSON_CreateObject();
-      cJSON * jAlternatives = cJSON_CreateArray();
-      cJSON * jStability = cJSON_CreateNumber(result.stability());
-      cJSON * jIsFinal = cJSON_CreateBool(result.is_final());
-      cJSON * jLanguageCode = cJSON_CreateString(result.language_code().c_str());
-      cJSON * jChannelTag = cJSON_CreateNumber(result.channel_tag());
-
-      auto duration = result.result_end_time();
-      int32_t seconds = duration.seconds();
-      int64_t nanos = duration.nanos();
-      int span = (int) trunc(seconds * 1000. + ((float) nanos / 1000000.));
-      cJSON * jResultEndTime = cJSON_CreateNumber(span);
-
-      cJSON_AddItemToObject(jResult, "stability", jStability);
-      cJSON_AddItemToObject(jResult, "is_final", jIsFinal);
-      cJSON_AddItemToObject(jResult, "alternatives", jAlternatives);
-      cJSON_AddItemToObject(jResult, "language_code", jLanguageCode);
-      cJSON_AddItemToObject(jResult, "channel_tag", jChannelTag);
-      cJSON_AddItemToObject(jResult, "result_end_time", jResultEndTime);
-
-      for (int a = 0; a < result.alternatives_size(); ++a) {
-        auto alternative = result.alternatives(a);
-        cJSON* jAlt = cJSON_CreateObject();
-        cJSON* jConfidence = cJSON_CreateNumber(alternative.confidence());
-        cJSON* jTranscript = cJSON_CreateString(alternative.transcript().c_str());
-        cJSON_AddItemToObject(jAlt, "confidence", jConfidence);
-        cJSON_AddItemToObject(jAlt, "transcript", jTranscript);
-
-        if (alternative.words_size() > 0) {
-          cJSON * jWords = cJSON_CreateArray();
-          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: %d words\n", alternative.words_size()) ;
-          for (int b = 0; b < alternative.words_size(); b++) {
-            auto words = alternative.words(b);
-            cJSON* jWord = cJSON_CreateObject();
-            cJSON_AddItemToObject(jWord, "word", cJSON_CreateString(words.word().c_str()));
-            if (words.has_start_time()) {
-              cJSON_AddItemToObject(jWord, "start_time", cJSON_CreateNumber(words.start_time().seconds()));
-            }
-            if (words.has_end_time()) {
-              cJSON_AddItemToObject(jWord, "end_time", cJSON_CreateNumber(words.end_time().seconds()));
-            }
-            int speaker_tag = words.speaker_tag();
-            if (speaker_tag > 0) {
-              cJSON_AddItemToObject(jWord, "speaker_tag", cJSON_CreateNumber(speaker_tag));
-            }
-            float confidence = words.confidence();
-            if (confidence > 0.0) {
-              cJSON_AddItemToObject(jWord, "confidence", cJSON_CreateNumber(confidence));
-            }
-
-            cJSON_AddItemToArray(jWords, jWord);
-          }
-          cJSON_AddItemToObject(jAlt, "words", jWords);
-        }
-        cJSON_AddItemToArray(jAlternatives, jAlt);
-      }
-
-      char* json = cJSON_PrintUnformatted(jResult);
-      cb->responseHandler(session, (const char *) json, cb->bugname);
-      free(json);
-
-      cJSON_Delete(jResult);
-    }
-
-    if (speech_event_type == StreamingRecognizeResponse_SpeechEventType_END_OF_SINGLE_UTTERANCE) {
-      // we only get this when we have requested it, and recognition stops after we get this
-      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got end_of_utterance\n") ;
-      cb->got_end_of_utterance = 1;
-      cb->responseHandler(session, "end_of_utterance", cb->bugname);
-      if (cb->wants_single_utterance) {
-        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: sending writesDone because we want only a single utterance\n") ;
-        streamer->writesDone();
-      }
-    }
-    switch_core_session_rwunlock(session);
-    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got %d responses\n", response.results_size());
-  }
-
-  {
-    switch_core_session_t* session = switch_core_session_locate(cb->sessionId);
-    if (session) {
-      grpc::Status status = streamer->finish();
-      if (11 == status.error_code()) {
-        if (std::string::npos != status.error_message().find("Exceeded maximum allowed stream duration")) {
-          cb->responseHandler(session, "max_duration_exceeded", cb->bugname);
-        }
-        else {
-          cb->responseHandler(session, "no_audio", cb->bugname);
-        }
-      }
-      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: finish() status %s (%d)\n", status.error_message().c_str(), status.error_code()) ;
-      switch_core_session_rwunlock(session);
-    }
-  }
-  return nullptr;
-}
+#include "google_glue.h"
+#include "generic_google_glue.h"
 
 extern "C" {
+	switch_status_t google_speech_init() {
+		const char* gcsServiceKeyFile = std::getenv("GOOGLE_APPLICATION_CREDENTIALS");
+		if (gcsServiceKeyFile) {
+		try {
+			auto creds = grpc::GoogleDefaultCredentials();
+		} catch (const std::exception& e) {
+			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, 
+			"Error initializing google api with provided credentials in %s: %s\n", gcsServiceKeyFile, e.what());
+			return SWITCH_STATUS_FALSE;
+		}
+		}
+		return SWITCH_STATUS_SUCCESS;
+	}
 
-    switch_status_t google_speech_init() {
-      const char* gcsServiceKeyFile = std::getenv("GOOGLE_APPLICATION_CREDENTIALS");
-      if (gcsServiceKeyFile) {
-        try {
-          auto creds = grpc::GoogleDefaultCredentials();
-        } catch (const std::exception& e) {
-          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, 
-            "Error initializing google api with provided credentials in %s: %s\n", gcsServiceKeyFile, e.what());
-          return SWITCH_STATUS_FALSE;
-        }
-      }
-      return SWITCH_STATUS_SUCCESS;
-    }
-
-    switch_status_t google_speech_cleanup() {
-      return SWITCH_STATUS_SUCCESS;
-    }
-    switch_status_t google_speech_session_init(switch_core_session_t *session, responseHandler_t responseHandler, 
-          uint32_t to_rate, uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char *bugname,
-          int single_utterance, int separate_recognition, int max_alternatives, int profanity_filter, int word_time_offset,
-          int punctuation, const char* model, int enhanced, const char* hints, char* play_file, void **ppUserData) {
-
-      switch_channel_t *channel = switch_core_session_get_channel(session);
-      auto read_codec = switch_core_session_get_read_codec(session);
-      uint32_t sampleRate = read_codec->implementation->actual_samples_per_second;
-      struct cap_cb *cb;
-      int err;
-
-      cb =(struct cap_cb *) switch_core_session_alloc(session, sizeof(*cb));
-      strncpy(cb->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
-      strncpy(cb->bugname, bugname, MAX_BUG_LEN);
-      cb->got_end_of_utterance = 0;
-      cb->wants_single_utterance = single_utterance;
-      if (play_file != NULL){
-        cb->play_file = 1;
-      }
-      
-      switch_mutex_init(&cb->mutex, SWITCH_MUTEX_NESTED, switch_core_session_get_pool(session));
-      if (sampleRate != to_rate) {
-          cb->resampler = speex_resampler_init(channels, sampleRate, to_rate, SWITCH_RESAMPLE_QUALITY, &err);
-        if (0 != err) {
-           switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing resampler: %s.\n",
-                                 switch_channel_get_name(channel), speex_resampler_strerror(err));
-          return SWITCH_STATUS_FALSE;
-        }
-      } else {
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "%s: no resampling needed for this call\n", switch_channel_get_name(channel));
-      }
-      cb->responseHandler = responseHandler;
-
-      // allocate vad if we are delaying connecting to the recognizer until we detect speech
-      if (switch_channel_var_true(channel, "START_RECOGNIZING_ON_VAD")) {
-        cb->vad = switch_vad_init(sampleRate, channels);
-        if (cb->vad) {
-          const char* var;
-          int mode = 2;
-          int silence_ms = 150;
-          int voice_ms = 250;
-          int debug = 0;
-
-          if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_MODE")) {
-            mode = atoi(var);
-          }
-          if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_SILENCE_MS")) {
-            silence_ms = atoi(var);
-          }
-          if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_VOICE_MS")) {
-            voice_ms = atoi(var);
-          }
-          if (var = switch_channel_get_variable(channel, "RECOGNIZER_VAD_VOICE_MS")) {
-            voice_ms = atoi(var);
-          }
-          switch_vad_set_mode(cb->vad, mode);
-          switch_vad_set_param(cb->vad, "silence_ms", silence_ms);
-          switch_vad_set_param(cb->vad, "voice_ms", voice_ms);
-          switch_vad_set_param(cb->vad, "debug", debug);
-        }
-      }
-
-      GStreamer *streamer = NULL;
-      try {
-        streamer = new GStreamer(session, channels, lang, interim, to_rate, sampleRate, single_utterance, separate_recognition, max_alternatives,
-         profanity_filter, word_time_offset, punctuation, model, enhanced, hints);
-        cb->streamer = streamer;
-      } catch (std::exception& e) {
-        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "%s: Error initializing gstreamer: %s.\n", 
-          switch_channel_get_name(channel), e.what());
-        return SWITCH_STATUS_FALSE;
-      }
-
-      if (!cb->vad) streamer->connect();
-
-      // create the read thread
-      switch_threadattr_t *thd_attr = NULL;
-      switch_memory_pool_t *pool = switch_core_session_get_pool(session);
-
-      switch_threadattr_create(&thd_attr, pool);
-      switch_threadattr_stacksize_set(thd_attr, SWITCH_THREAD_STACKSIZE);
-      switch_thread_create(&cb->thread, thd_attr, grpc_read_thread, cb, pool);
-
-      *ppUserData = cb;
-      return SWITCH_STATUS_SUCCESS;
-    }
-
-    switch_status_t google_speech_session_cleanup(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug) {
-      switch_channel_t *channel = switch_core_session_get_channel(session);
-
-      if (bug) {
-        struct cap_cb *cb = (struct cap_cb *) switch_core_media_bug_get_user_data(bug);
-        switch_mutex_lock(cb->mutex);
-
-        if (!switch_channel_get_private(channel, cb->bugname)) {
-          // race condition
-          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s Bug is not attached (race).\n", switch_channel_get_name(channel));
-          switch_mutex_unlock(cb->mutex);
-          return SWITCH_STATUS_FALSE;
-        }
-        switch_channel_set_private(channel, cb->bugname, NULL);
-
-      // stop playback if available
-       if (cb->play_file == 1){ 
-          if (switch_channel_test_flag(channel, CF_BROADCAST)) {
-		        switch_channel_stop_broadcast(channel);
-	        } else {
-		        switch_channel_set_flag_value(channel, CF_BREAK, 1);
-        	}
-        }
-
-        // close connection and get final responses
-        GStreamer* streamer = (GStreamer *) cb->streamer;
-
-        if (streamer) {
-          streamer->writesDone();
-
-          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "google_speech_session_cleanup: GStreamer (%p) waiting for read thread to complete\n", (void*)streamer);
-          switch_status_t st;
-          switch_thread_join(&st, cb->thread);
-          switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "google_speech_session_cleanup:  GStreamer (%p) read thread completed\n", (void*)streamer);
-
-          delete streamer;
-          cb->streamer = NULL;
-        }
-
-        if (cb->resampler) {
-          speex_resampler_destroy(cb->resampler);
-        }
-        if (cb->vad) {
-          switch_vad_destroy(&cb->vad);
-          cb->vad = nullptr;
-        }
-        if (!channelIsClosing) {
-          switch_core_media_bug_remove(session, &bug);
-        }
-
-			  switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "google_speech_session_cleanup: Closed stream\n");
-
-			  switch_mutex_unlock(cb->mutex);
-
-
-			  return SWITCH_STATUS_SUCCESS;
-      }
-
-      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s Bug is not attached.\n", switch_channel_get_name(channel));
-      return SWITCH_STATUS_FALSE;
-    }
-
-    switch_bool_t google_speech_frame(switch_media_bug_t *bug, void* user_data) {
-    	switch_core_session_t *session = switch_core_media_bug_get_session(bug);
-    	struct cap_cb *cb = (struct cap_cb *) user_data;
-		  if (cb->streamer && (!cb->wants_single_utterance || !cb->got_end_of_utterance)) {
-        GStreamer* streamer = (GStreamer *) cb->streamer;
-        uint8_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
-        switch_frame_t frame = {};
-        frame.data = data;
-        frame.buflen = SWITCH_RECOMMENDED_BUFFER_SIZE;
-
-        if (switch_mutex_trylock(cb->mutex) == SWITCH_STATUS_SUCCESS) {
-          while (switch_core_media_bug_read(bug, &frame, SWITCH_TRUE) == SWITCH_STATUS_SUCCESS && !switch_test_flag((&frame), SFF_CNG)) {
-            if (frame.datalen) {
-              if (cb->vad && !streamer->isConnected()) {
-                switch_vad_state_t state = switch_vad_process(cb->vad, (int16_t*) frame.data, frame.samples);
-                if (state == SWITCH_VAD_STATE_START_TALKING) {
-                  switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "detected speech, connect to google speech now\n");
-                  streamer->connect();
-                  cb->responseHandler(session, "vad_detected", cb->bugname);
-                }
-              }
-
-              if (cb->resampler) {
-                spx_int16_t out[SWITCH_RECOMMENDED_BUFFER_SIZE];
-                spx_uint32_t out_len = SWITCH_RECOMMENDED_BUFFER_SIZE;
-                spx_uint32_t in_len = frame.samples;
-                size_t written;
-
-                speex_resampler_process_interleaved_int(cb->resampler,
-                  (const spx_int16_t *) frame.data,
-                  (spx_uint32_t *) &in_len,
-                  &out[0],
-                  &out_len);
-                streamer->write( &out[0], sizeof(spx_int16_t) * out_len);
-              }
-              else {
-                streamer->write( frame.data, sizeof(spx_int16_t) * frame.samples);
-              }
-            }
-          }
-          switch_mutex_unlock(cb->mutex);
-        }
-      }
-      return SWITCH_TRUE;
-    }
+	switch_status_t google_speech_cleanup() {
+		return SWITCH_STATUS_SUCCESS;
+	}
 }

mod_google_transcribe/google_glue.h

@@ -1,13 +1,28 @@
 #ifndef __GOOGLE_GLUE_H__
 #define __GOOGLE_GLUE_H__
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+
 switch_status_t google_speech_init();
 switch_status_t google_speech_cleanup();
-switch_status_t google_speech_session_init(switch_core_session_t *session, responseHandler_t responseHandler, 
+switch_status_t google_speech_session_init_v1(switch_core_session_t *session, responseHandler_t responseHandler, 
 		uint32_t to_rate, uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char *bugname, int single_utterence,
-		int separate_recognition, int max_alternatives, int profinity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, 
+		int separate_recognition, int max_alternatives, int profanity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, 
 		const char* hints, char* play_file, void **ppUserData);
-switch_status_t google_speech_session_cleanup(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug);
-switch_bool_t google_speech_frame(switch_media_bug_t *bug, void* user_data);
+switch_status_t google_speech_session_cleanup_v1(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug);
+switch_status_t google_speech_session_init_v2(switch_core_session_t *session, responseHandler_t responseHandler, 
+		uint32_t to_rate, uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char *bugname, int single_utterence,
+		int separate_recognition, int max_alternatives, int profanity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, 
+		const char* hints, char* play_file, void **ppUserData);
+switch_status_t google_speech_session_cleanup_v2(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug);
+
+switch_bool_t google_speech_frame_v1(switch_media_bug_t *bug, void* user_data);
+switch_bool_t google_speech_frame_v2(switch_media_bug_t *bug, void* user_data);
+
+#ifdef __cplusplus
+}
+#endif
 
 #endif

mod_google_transcribe/google_glue_v1.cpp

@@ -0,0 +1,397 @@
+#include <switch.h>
+#include <switch_json.h>
+#include <grpc++/grpc++.h>
+
+#include "mod_google_transcribe.h"
+
+#include "gstreamer.h"
+#include "generic_google_glue.h"
+
+#include "google/cloud/speech/v1p1beta1/cloud_speech.grpc.pb.h"
+
+using google::cloud::speech::v1p1beta1::RecognitionConfig;
+using google::cloud::speech::v1p1beta1::Speech;
+using google::cloud::speech::v1p1beta1::SpeechContext;
+using google::cloud::speech::v1p1beta1::StreamingRecognizeRequest;
+using google::cloud::speech::v1p1beta1::StreamingRecognizeResponse;
+using google::cloud::speech::v1p1beta1::SpeakerDiarizationConfig;
+using google::cloud::speech::v1p1beta1::SpeechAdaptation;
+using google::cloud::speech::v1p1beta1::PhraseSet;
+using google::cloud::speech::v1p1beta1::PhraseSet_Phrase;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_DISCUSSION;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_PRESENTATION;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_PHONE_CALL;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_VOICEMAIL;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_PROFESSIONALLY_PRODUCED;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_VOICE_SEARCH;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_VOICE_COMMAND;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_InteractionType_DICTATION;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_MicrophoneDistance_NEARFIELD;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_MicrophoneDistance_MIDFIELD;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_MicrophoneDistance_FARFIELD;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_OriginalMediaType_AUDIO;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_OriginalMediaType_VIDEO;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_SMARTPHONE;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_PC;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_PHONE_LINE;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_VEHICLE;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_OTHER_OUTDOOR_DEVICE;
+using google::cloud::speech::v1p1beta1::RecognitionMetadata_RecordingDeviceType_OTHER_INDOOR_DEVICE;
+using google::cloud::speech::v1p1beta1::StreamingRecognizeResponse_SpeechEventType_END_OF_SINGLE_UTTERANCE;
+using google::rpc::Status;
+
+typedef GStreamer<StreamingRecognizeRequest, StreamingRecognizeResponse, Speech::Stub> GStreamer_V1;
+
+template<>
+GStreamer<StreamingRecognizeRequest, StreamingRecognizeResponse, Speech::Stub>::GStreamer(
+    switch_core_session_t *session, 
+    uint32_t channels, 
+    char* lang, 
+    int interim, 
+    uint32_t config_sample_rate,
+		uint32_t samples_per_second,
+    int single_utterance, 
+    int separate_recognition,
+		int max_alternatives, 
+    int profanity_filter, 
+    int word_time_offset, 
+    int punctuation, 
+    const char* model, 
+    int enhanced, 
+		const char* hints) : m_session(session), m_writesDone(false), m_connected(false), 
+      m_audioBuffer(CHUNKSIZE, 15) {
+  
+    switch_channel_t *channel = switch_core_session_get_channel(session);
+    m_channel = create_grpc_channel(channel);
+  	m_stub = Speech::NewStub(m_channel);
+  		
+		auto* streaming_config = m_request.mutable_streaming_config();
+		RecognitionConfig* config = streaming_config->mutable_config();
+
+    streaming_config->set_interim_results(interim);
+    if (single_utterance == 1) {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_single_utterance\n");
+      streaming_config->set_single_utterance(true);
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_single_utterance is FALSE\n");
+      streaming_config->set_single_utterance(false);
+    }
+
+		config->set_language_code(lang);
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "transcribe language %s \n", lang);
+    
+  	config->set_sample_rate_hertz(config_sample_rate);
+
+		config->set_encoding(RecognitionConfig::LINEAR16);
+
+    // the rest of config comes from channel vars
+
+    // number of channels in the audio stream (default: 1)
+    if (channels > 1) {
+      config->set_audio_channel_count(channels);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "audio_channel_count %d\n", channels);
+
+      // transcribe each separately?
+      if (separate_recognition == 1) {
+        config->set_enable_separate_recognition_per_channel(true);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_separate_recognition_per_channel on\n");
+      }
+    }
+
+    // max alternatives
+    if (max_alternatives > 1) {
+      config->set_max_alternatives(max_alternatives);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "max_alternatives %d\n", max_alternatives);
+    }
+
+    // profanity filter
+    if (profanity_filter == 1) {
+      config->set_profanity_filter(true);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "profanity_filter\n");
+    }
+
+    // enable word offsets
+    if (word_time_offset == 1) {
+      config->set_enable_word_time_offsets(true);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_word_time_offsets\n");
+    }
+
+    // enable automatic punctuation
+    if (punctuation == 1) {
+      config->set_enable_automatic_punctuation(true);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_automatic_punctuation\n");
+    }
+    else {
+      config->set_enable_automatic_punctuation(false);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "disable_automatic_punctuation\n");
+    }
+
+    // speech model
+    if (model != NULL) {
+      config->set_model(model);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "speech model %s\n", model);
+    }
+
+    // use enhanced model
+    if (enhanced == 1) {
+      config->set_use_enhanced(true);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "use_enhanced\n");
+    }
+
+    // hints  
+    if (hints != NULL) {
+      auto* adaptation = config->mutable_adaptation();
+      auto* phrase_set = adaptation->add_phrase_sets();
+      google_speech_configure_grammar_hints(m_session, channel, hints, phrase_set);
+    }
+
+    const char* var;
+    
+    // alternative language
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_ALTERNATIVE_LANGUAGE_CODES")) {
+      char *alt_langs[3] = { 0 };
+      int argc = switch_separate_string((char *) var, ',', alt_langs, 3);
+      for (int i = 0; i < argc; i++) {
+        config->add_alternative_language_codes(alt_langs[i]);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "added alternative lang %s\n", alt_langs[i]);
+      }
+    }
+
+    // speaker diarization
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION")) {
+      auto* diarization_config = config->mutable_diarization_config();
+      diarization_config->set_enable_speaker_diarization(true);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enabling speaker diarization\n", var);
+      if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION_MIN_SPEAKER_COUNT")) {
+        int count = std::max(atoi(var), 1);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting min speaker count to %d\n", count);
+        diarization_config->set_min_speaker_count(count);
+      }
+      if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION_MAX_SPEAKER_COUNT")) {
+        int count = std::max(atoi(var), 2);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting max speaker count to %d\n", count);
+        diarization_config->set_max_speaker_count(count);
+      }
+    }
+
+    // recognition metadata
+    auto* metadata = config->mutable_metadata();
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_INTERACTION_TYPE")) {
+      if (case_insensitive_match("discussion", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_DISCUSSION);
+      if (case_insensitive_match("presentation", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_PRESENTATION);
+      if (case_insensitive_match("phone_call", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_PHONE_CALL);
+      if (case_insensitive_match("voicemail", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_VOICEMAIL);
+      if (case_insensitive_match("professionally_produced", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_PROFESSIONALLY_PRODUCED);
+      if (case_insensitive_match("voice_search", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_VOICE_SEARCH);
+      if (case_insensitive_match("voice_command", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_VOICE_COMMAND);
+      if (case_insensitive_match("dictation", var)) metadata->set_interaction_type(RecognitionMetadata_InteractionType_DICTATION);
+    }
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_INDUSTRY_NAICS_CODE")) {
+      metadata->set_industry_naics_code_of_audio(atoi(var));
+    }
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_MICROPHONE_DISTANCE")) {
+      if (case_insensitive_match("nearfield", var)) metadata->set_microphone_distance(RecognitionMetadata_MicrophoneDistance_NEARFIELD);
+      if (case_insensitive_match("midfield", var)) metadata->set_microphone_distance(RecognitionMetadata_MicrophoneDistance_MIDFIELD);
+      if (case_insensitive_match("farfield", var)) metadata->set_microphone_distance(RecognitionMetadata_MicrophoneDistance_FARFIELD);
+    }
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_ORIGINAL_MEDIA_TYPE")) {
+      if (case_insensitive_match("audio", var)) metadata->set_original_media_type(RecognitionMetadata_OriginalMediaType_AUDIO);
+      if (case_insensitive_match("video", var)) metadata->set_original_media_type(RecognitionMetadata_OriginalMediaType_VIDEO);
+    }
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_METADATA_RECORDING_DEVICE_TYPE")) {
+      if (case_insensitive_match("smartphone", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_SMARTPHONE);
+      if (case_insensitive_match("pc", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_PC);
+      if (case_insensitive_match("phone_line", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_PHONE_LINE);
+      if (case_insensitive_match("vehicle", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_VEHICLE);
+      if (case_insensitive_match("other_outdoor_device", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_OTHER_OUTDOOR_DEVICE);
+      if (case_insensitive_match("other_indoor_device", var)) metadata->set_recording_device_type(RecognitionMetadata_RecordingDeviceType_OTHER_INDOOR_DEVICE);
+    }
+}
+
+static void *SWITCH_THREAD_FUNC grpc_read_thread(switch_thread_t *thread, void *obj) {
+  static int count;
+	struct cap_cb *cb = (struct cap_cb *) obj;
+	GStreamer_V1* streamer = (GStreamer_V1 *) cb->streamer;
+
+  bool connected = streamer->waitForConnect();
+  if (!connected) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "google transcribe grpc read thread exiting since we didn't connect\n") ;
+    return nullptr;
+  }
+
+  // Read responses.
+  StreamingRecognizeResponse response;
+  while (streamer->read(&response)) {  // Returns false when no more to read.
+    switch_core_session_t* session = switch_core_session_locate(cb->sessionId);
+    if (!session) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "grpc_read_thread: session %s is gone!\n", cb->sessionId) ;
+      return nullptr;
+    }
+    count++;
+    auto speech_event_type = response.speech_event_type();
+    if (response.has_error()) {
+      Status status = response.error();
+      //error 11 is handled in finished session, avoid sending jambonz_transcribe::error event for this here.
+      if (11 == status.code()) {
+        continue;
+      }
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "grpc_read_thread: error %s (%d)\n", status.message().c_str(), status.code()) ;
+      cJSON* json = cJSON_CreateObject();
+      cJSON_AddStringToObject(json, "type", "error");
+      cJSON_AddStringToObject(json, "error_cause", "stream_response");
+      cJSON_AddStringToObject(json, "error", status.message().c_str());
+      char* jsonString = cJSON_PrintUnformatted(json);
+      cb->responseHandler(session, jsonString, cb->bugname);
+      free(jsonString);
+      cJSON_Delete(json);
+    }
+    
+    if (cb->play_file == 1) {
+      cb->responseHandler(session, "play_interrupt", cb->bugname);
+    }
+    
+    for (int r = 0; r < response.results_size(); ++r) {
+      auto result = response.results(r);
+      cJSON * jResult = cJSON_CreateObject();
+      cJSON * jAlternatives = cJSON_CreateArray();
+      cJSON * jStability = cJSON_CreateNumber(result.stability());
+      cJSON * jIsFinal = cJSON_CreateBool(result.is_final());
+      cJSON * jLanguageCode = cJSON_CreateString(result.language_code().c_str());
+      cJSON * jChannelTag = cJSON_CreateNumber(result.channel_tag());
+
+      auto duration = result.result_end_time();
+      int32_t seconds = duration.seconds();
+      int64_t nanos = duration.nanos();
+      int span = (int) trunc(seconds * 1000. + ((float) nanos / 1000000.));
+      cJSON * jResultEndTime = cJSON_CreateNumber(span);
+
+      cJSON_AddItemToObject(jResult, "stability", jStability);
+      cJSON_AddItemToObject(jResult, "is_final", jIsFinal);
+      cJSON_AddItemToObject(jResult, "alternatives", jAlternatives);
+      cJSON_AddItemToObject(jResult, "language_code", jLanguageCode);
+      cJSON_AddItemToObject(jResult, "channel_tag", jChannelTag);
+      cJSON_AddItemToObject(jResult, "result_end_time", jResultEndTime);
+
+      for (int a = 0; a < result.alternatives_size(); ++a) {
+        auto alternative = result.alternatives(a);
+        cJSON* jAlt = cJSON_CreateObject();
+        cJSON* jConfidence = cJSON_CreateNumber(alternative.confidence());
+        cJSON* jTranscript = cJSON_CreateString(alternative.transcript().c_str());
+        cJSON_AddItemToObject(jAlt, "confidence", jConfidence);
+        cJSON_AddItemToObject(jAlt, "transcript", jTranscript);
+
+        if (alternative.words_size() > 0) {
+          cJSON * jWords = cJSON_CreateArray();
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: %d words\n", alternative.words_size()) ;
+          for (int b = 0; b < alternative.words_size(); b++) {
+            auto words = alternative.words(b);
+            cJSON* jWord = cJSON_CreateObject();
+            cJSON_AddItemToObject(jWord, "word", cJSON_CreateString(words.word().c_str()));
+            if (words.has_start_time()) {
+              cJSON_AddItemToObject(jWord, "start_time", cJSON_CreateNumber(words.start_time().seconds()));
+            }
+            if (words.has_end_time()) {
+              cJSON_AddItemToObject(jWord, "end_time", cJSON_CreateNumber(words.end_time().seconds()));
+            }
+            int speaker_tag = words.speaker_tag();
+            if (speaker_tag > 0) {
+              cJSON_AddItemToObject(jWord, "speaker_tag", cJSON_CreateNumber(speaker_tag));
+            }
+            float confidence = words.confidence();
+            if (confidence > 0.0) {
+              cJSON_AddItemToObject(jWord, "confidence", cJSON_CreateNumber(confidence));
+            }
+
+            cJSON_AddItemToArray(jWords, jWord);
+          }
+          cJSON_AddItemToObject(jAlt, "words", jWords);
+        }
+        cJSON_AddItemToArray(jAlternatives, jAlt);
+      }
+
+      char* json = cJSON_PrintUnformatted(jResult);
+      cb->responseHandler(session, (const char *) json, cb->bugname);
+      free(json);
+
+      cJSON_Delete(jResult);
+    }
+
+    if (speech_event_type == StreamingRecognizeResponse_SpeechEventType_END_OF_SINGLE_UTTERANCE) {
+      // we only get this when we have requested it, and recognition stops after we get this
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got end_of_utterance\n") ;
+      cb->got_end_of_utterance = 1;
+      cb->responseHandler(session, "end_of_utterance", cb->bugname);
+      if (cb->wants_single_utterance) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: sending writesDone because we want only a single utterance\n") ;
+        streamer->writesDone();
+      }
+    }
+    switch_core_session_rwunlock(session);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got %d responses\n", response.results_size());
+  }
+
+  {
+    switch_core_session_t* session = switch_core_session_locate(cb->sessionId);
+    if (session) {
+      grpc::Status status = streamer->finish();
+      if (11 == status.error_code()) {
+        if (std::string::npos != status.error_message().find("Exceeded maximum allowed stream duration")) {
+          cb->responseHandler(session, "max_duration_exceeded", cb->bugname);
+        }
+        else {
+          cb->responseHandler(session, "no_audio", cb->bugname);
+        }
+      }
+      else if (status.error_code() != 0) {
+        cJSON* json = cJSON_CreateObject();
+        cJSON_AddStringToObject(json, "type", "error");
+        cJSON_AddStringToObject(json, "error_cause", "stream_close");
+        cJSON_AddItemToObject(json, "error_code", cJSON_CreateNumber(status.error_code()));
+        cJSON_AddStringToObject(json, "error_message", status.error_message().c_str());
+        char* jsonString = cJSON_PrintUnformatted(json);
+        cb->responseHandler(session, jsonString, cb->bugname);
+        free(jsonString);
+        cJSON_Delete(json);
+      }
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: finish() status %s (%d)\n", status.error_message().c_str(), status.error_code()) ;
+      switch_core_session_rwunlock(session);
+    }
+  }
+  return nullptr;
+}
+
+template<>
+bool GStreamer<StreamingRecognizeRequest, StreamingRecognizeResponse, Speech::Stub>::write(void* data, uint32_t datalen) {
+  if (!m_connected) {
+    if (datalen % CHUNKSIZE == 0) {
+      m_audioBuffer.add(data, datalen);
+    }
+    return true;
+  }
+  m_request.set_audio_content(data, datalen);
+  bool ok = m_streamer->Write(m_request);
+  return ok;
+}
+
+extern "C" {
+
+    switch_status_t google_speech_session_cleanup_v1(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug) {
+      return google_speech_session_cleanup<GStreamer_V1>(session, channelIsClosing, bug);
+    }
+
+    switch_bool_t google_speech_frame_v1(switch_media_bug_t *bug, void* user_data) {
+      return google_speech_frame<GStreamer_V1>(bug, user_data);
+    }
+
+    switch_status_t google_speech_session_init_v1(switch_core_session_t *session, responseHandler_t responseHandler,
+		  uint32_t to_rate, uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char *bugname, int single_utterance,
+		  int separate_recognition, int max_alternatives, int profanity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, 
+		  const char* hints, char* play_file, void **ppUserData) {
+      return google_speech_session_init<GStreamer_V1>(session, responseHandler, grpc_read_thread, to_rate, samples_per_second, channels,
+        lang, interim, bugname, single_utterance, separate_recognition, max_alternatives, profanity_filter,
+        word_time_offset, punctuation, model, enhanced, hints, play_file, ppUserData);
+    }
+
+}

mod_google_transcribe/google_glue_v2.cpp

@@ -0,0 +1,400 @@
+#include <switch.h>
+#include <switch_json.h>
+#include <grpc++/grpc++.h>
+
+#include "mod_google_transcribe.h"
+#include "gstreamer.h"
+#include "generic_google_glue.h"
+
+#include "google/cloud/speech/v2/cloud_speech.grpc.pb.h"
+
+using google::cloud::speech::v2::RecognitionConfig;
+using google::cloud::speech::v2::Speech;
+using google::cloud::speech::v2::StreamingRecognizeRequest;
+using google::cloud::speech::v2::StreamingRecognizeResponse;
+using google::cloud::speech::v2::SpeakerDiarizationConfig;
+using google::cloud::speech::v2::SpeechAdaptation;
+using google::cloud::speech::v2::SpeechRecognitionAlternative;
+using google::cloud::speech::v2::PhraseSet;
+using google::cloud::speech::v2::PhraseSet_Phrase;
+using google::cloud::speech::v2::StreamingRecognizeResponse_SpeechEventType_END_OF_SINGLE_UTTERANCE;
+using google::cloud::speech::v2::StreamingRecognizeResponse_SpeechEventType_SPEECH_ACTIVITY_BEGIN;
+using google::cloud::speech::v2::StreamingRecognizeResponse_SpeechEventType_SPEECH_ACTIVITY_END;
+using google::cloud::speech::v2::ExplicitDecodingConfig_AudioEncoding_LINEAR16;
+using google::cloud::speech::v2::RecognitionFeatures_MultiChannelMode_SEPARATE_RECOGNITION_PER_CHANNEL;
+using google::cloud::speech::v2::SpeechAdaptation_AdaptationPhraseSet;
+using google::rpc::Status;
+
+typedef GStreamer<StreamingRecognizeRequest, StreamingRecognizeResponse, Speech::Stub> GStreamer_V2;
+
+template<>
+GStreamer<StreamingRecognizeRequest, StreamingRecognizeResponse, Speech::Stub>::GStreamer(
+    switch_core_session_t *session, 
+    uint32_t channels, 
+    char* lang, 
+    int interim, 
+    uint32_t config_sample_rate,
+	uint32_t samples_per_second,
+    int single_utterance, 
+    int separate_recognition,
+	int max_alternatives, 
+    int profanity_filter, 
+    int word_time_offset, 
+    int punctuation, 
+    const char* model, 
+    int enhanced, 
+	const char* hints) : m_session(session), m_writesDone(false), m_connected(false),
+    m_audioBuffer(CHUNKSIZE, 15) {
+  
+    switch_channel_t *channel = switch_core_session_get_channel(session);
+    m_channel = create_grpc_channel(channel);
+  	m_stub = Speech::NewStub(m_channel);
+
+	auto streaming_config = m_request.mutable_streaming_config();
+    const char* var;
+
+    // The parent of the recognizer must still be provided even if the wildcard
+    // recognizer is used rather than a pre-prepared recognizer.
+    std::string recognizer;
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_RECOGNIZER_PARENT")) {
+        recognizer = var;
+        recognizer += "/recognizers/";
+    } else {
+        throw std::runtime_error("The v2 Speech-To-Text library requires GOOGLE_SPEECH_RECOGNIZER_PARENT to be set");
+    }
+
+    // Use the recognizer specified in the variable or just use the wildcard if this is not set.
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_RECOGNIZER_ID")) {
+        recognizer += var;
+    } else {
+        recognizer += "_";
+
+        RecognitionConfig* config = streaming_config->mutable_config();
+        config->add_language_codes(lang);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "transcribe language %s\n", lang);
+        
+        // alternative language
+        if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_ALTERNATIVE_LANGUAGE_CODES")) {
+            char *alt_langs[3] = { 0 };
+            int argc = switch_separate_string((char *) var, ',', alt_langs, 3);
+            for (int i = 0; i < argc; i++) {
+                config->add_language_codes(alt_langs[i]);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "added alternative lang %s\n", alt_langs[i]);
+            }
+        }
+
+        config->mutable_explicit_decoding_config()->set_sample_rate_hertz(config_sample_rate);
+        config->mutable_explicit_decoding_config()->set_encoding(ExplicitDecodingConfig_AudioEncoding_LINEAR16);
+
+        // number of channels in the audio stream (default: 1)
+        // N.B. It is essential to set this configuration value in v2 even if it doesn't deviate from the default.
+        config->mutable_explicit_decoding_config()->set_audio_channel_count(channels);
+        switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "audio_channel_count %d\n", channels);
+        if (channels > 1) {
+            // transcribe each separately?
+            if (separate_recognition == 1) {
+                config->mutable_features()->set_multi_channel_mode(RecognitionFeatures_MultiChannelMode_SEPARATE_RECOGNITION_PER_CHANNEL);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_separate_recognition_per_channel on\n");
+            }
+        }
+
+        // max alternatives
+        if (max_alternatives > 1) {
+            config->mutable_features()->set_max_alternatives(max_alternatives);
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "max_alternatives %d\n", max_alternatives);
+        }
+
+        // profanity filter
+        if (profanity_filter == 1) {
+            config->mutable_features()->set_profanity_filter(true);
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "profanity_filter\n");
+        }
+
+        // enable word offsets
+        if (word_time_offset == 1) {
+            config->mutable_features()->set_enable_word_time_offsets(true);
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_word_time_offsets\n");
+        }
+
+        // enable automatic punctuation
+        if (punctuation == 1) {
+            config->mutable_features()->set_enable_automatic_punctuation(true);
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enable_automatic_punctuation\n");
+        }
+        else {
+            config->mutable_features()->set_enable_automatic_punctuation(false);
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "disable_automatic_punctuation\n");
+        }
+
+        // speech model
+        if (model != NULL) {
+            config->set_model(model);
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "speech model %s\n", model);
+        }
+
+        // hints  
+        if (hints != NULL) {
+            auto* adaptation = config->mutable_adaptation();
+            auto* phrase_set = adaptation->add_phrase_sets()->mutable_inline_phrase_set();
+            google_speech_configure_grammar_hints(m_session, channel, hints, phrase_set);
+        }
+
+        // the rest of config comes from channel vars
+
+        // speaker diarization
+        // N.B. At the moment there does not seem to be any combination of model, language and location which supports diarization for STT v2.
+        // See https://stackoverflow.com/questions/76779418/speaker-diarization-is-disabled-even-for-supported-languages-in-google-speech-to
+        if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION")) {
+            auto* diarization_config = config->mutable_features()->mutable_diarization_config();
+            // There is no enable function in v2
+            // diarization_config->set_enable_speaker_diarization(true);
+            // switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "enabling speaker diarization\n", var);
+            if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION_MIN_SPEAKER_COUNT")) {
+                int count = std::max(atoi(var), 1);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting min speaker count to %d\n", count);
+                diarization_config->set_min_speaker_count(count);
+            }
+            if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SPEAKER_DIARIZATION_MAX_SPEAKER_COUNT")) {
+                int count = std::max(atoi(var), 2);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting max speaker count to %d\n", count);
+                diarization_config->set_max_speaker_count(count);
+            }
+        }
+        if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_TRANSCRIPTION_NORMALIZATION")) {
+          // parse JSON string
+        cJSON *json_array = cJSON_Parse(var);
+
+        int array_size = cJSON_GetArraySize(json_array);
+
+        for(int i=0; i<array_size; i++) {
+            cJSON* json_item = cJSON_GetArrayItem(json_array, i);
+
+            auto entry = config->mutable_transcript_normalization()->add_entries();
+
+            std::string search_string = cJSON_GetObjectItem(json_item, "search")->valuestring;
+            std::string replacement_string = cJSON_GetObjectItem(json_item, "replace")->valuestring;
+            bool case_sensitive = cJSON_GetObjectItem(json_item, "case_sensitive")->valueint != 0;
+
+            entry->set_search(search_string);
+            entry->set_replace(replacement_string);
+            entry->set_case_sensitive(case_sensitive);
+
+            switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG,
+              "TRANSCRIPTION_NORMALIZATION search %s, replace %s, set_case_sensitive %d\n", search_string.c_str(), replacement_string.c_str(), case_sensitive);
+        }
+        // clean json
+        cJSON_Delete(json_array);
+      }
+    }
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_START_TIMEOUT_MS")) {
+      auto ms = atoi(var);
+      streaming_config->mutable_streaming_features()->mutable_voice_activity_timeout()->mutable_speech_start_timeout()->set_nanos(ms * 1000000);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting speech_start_timeout to %d milliseconds\n", ms);
+    }
+
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_END_TIMEOUT_MS")) {
+      auto ms = atoi(var);
+      streaming_config->mutable_streaming_features()->mutable_voice_activity_timeout()->mutable_speech_end_timeout()->set_nanos(ms * 1000000);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting speech_end_timeout to %d milliseconds\n", ms);
+    }
+
+    if (var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_ENABLE_VOICE_ACTIVITY_EVENTS")) {
+      bool enabled = !strcmp(var, "true") ? 1 : 0;
+      streaming_config->mutable_streaming_features()->set_enable_voice_activity_events(enabled);
+      switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "setting enable_voice_activity_events to %d \n", enabled);
+    }
+
+    m_request.set_recognizer(recognizer);
+    switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_DEBUG, "using recognizer: %s\n", recognizer.c_str());
+
+    // This must be set whether a recognizer id is provided or not, because it cannot be configured as part of a recognizer.
+    if (interim > 0) {
+        streaming_config->mutable_streaming_features()->set_interim_results(interim > 0);
+    }
+}
+
+static void *SWITCH_THREAD_FUNC grpc_read_thread(switch_thread_t *thread, void *obj) {
+  static int count;
+	struct cap_cb *cb = (struct cap_cb *) obj;
+	GStreamer_V2* streamer = (GStreamer_V2 *) cb->streamer;
+
+  bool connected = streamer->waitForConnect();
+  if (!connected) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "google transcribe grpc read thread exiting since we didn't connect\n") ;
+    return nullptr;
+  }
+
+  // Read responses.
+  StreamingRecognizeResponse response;
+  while (streamer->read(&response)) {  // Returns false when no more to read.
+    switch_core_session_t* session = switch_core_session_locate(cb->sessionId);
+    if (!session) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "grpc_read_thread: session %s is gone!\n", cb->sessionId) ;
+      return nullptr;
+    }
+    count++;
+    
+    if (cb->play_file == 1){
+      cb->responseHandler(session, "play_interrupt", cb->bugname);
+    }
+    
+    for (int r = 0; r < response.results_size(); ++r) {
+      auto result = response.results(r);
+      cJSON * jResult = cJSON_CreateObject();
+      cJSON * jAlternatives = cJSON_CreateArray();
+      cJSON * jStability = cJSON_CreateNumber(result.stability());
+      cJSON * jIsFinal = cJSON_CreateBool(result.is_final());
+      cJSON * jLanguageCode = cJSON_CreateString(result.language_code().c_str());
+      cJSON * jChannelTag = cJSON_CreateNumber(result.channel_tag());
+
+      auto duration = result.result_end_offset();
+      int32_t seconds = duration.seconds();
+      int64_t nanos = duration.nanos();
+      int span = (int) trunc(seconds * 1000. + ((float) nanos / 1000000.));
+      cJSON * jResultEndTime = cJSON_CreateNumber(span);
+
+      cJSON_AddItemToObject(jResult, "stability", jStability);
+      cJSON_AddItemToObject(jResult, "is_final", jIsFinal);
+      cJSON_AddItemToObject(jResult, "alternatives", jAlternatives);
+      cJSON_AddItemToObject(jResult, "language_code", jLanguageCode);
+      cJSON_AddItemToObject(jResult, "channel_tag", jChannelTag);
+      cJSON_AddItemToObject(jResult, "result_end_time", jResultEndTime);
+
+      if (result.alternatives_size() == 0) {
+        SpeechRecognitionAlternative alternative;
+        alternative.set_confidence(0.0);
+        alternative.set_transcript("");
+        *result.add_alternatives() = alternative;
+      }
+      for (int a = 0; a < result.alternatives_size(); ++a) {
+        auto alternative = result.alternatives(a);
+        cJSON* jAlt = cJSON_CreateObject();
+        cJSON* jConfidence = cJSON_CreateNumber(alternative.confidence());
+        cJSON* jTranscript = cJSON_CreateString(alternative.transcript().c_str());
+        cJSON_AddItemToObject(jAlt, "confidence", jConfidence);
+        cJSON_AddItemToObject(jAlt, "transcript", jTranscript);
+
+        if (alternative.words_size() > 0) {
+          cJSON * jWords = cJSON_CreateArray();
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: %d words\n", alternative.words_size()) ;
+          for (int b = 0; b < alternative.words_size(); b++) {
+            auto words = alternative.words(b);
+            cJSON* jWord = cJSON_CreateObject();
+            cJSON_AddItemToObject(jWord, "word", cJSON_CreateString(words.word().c_str()));
+            if (words.has_start_offset()) {
+              cJSON_AddItemToObject(jWord, "start_offset", cJSON_CreateNumber(words.start_offset().seconds()));
+            }
+            if (words.has_end_offset()) {
+              cJSON_AddItemToObject(jWord, "end_offset", cJSON_CreateNumber(words.end_offset().seconds()));
+            }
+            auto speaker_label = words.speaker_label();
+            if (speaker_label.size() > 0) {
+              cJSON_AddItemToObject(jWord, "speaker_label", cJSON_CreateString(speaker_label.c_str()));
+            }
+            float confidence = words.confidence();
+            if (confidence > 0.0) {
+              cJSON_AddItemToObject(jWord, "confidence", cJSON_CreateNumber(confidence));
+            }
+
+            cJSON_AddItemToArray(jWords, jWord);
+          }
+          cJSON_AddItemToObject(jAlt, "words", jWords);
+        }
+        cJSON_AddItemToArray(jAlternatives, jAlt);
+      }
+
+      char* json = cJSON_PrintUnformatted(jResult);
+      cb->responseHandler(session, (const char *) json, cb->bugname);
+      free(json);
+
+      cJSON_Delete(jResult);
+    }
+
+    auto speech_event_type = response.speech_event_type();
+    if (speech_event_type == StreamingRecognizeResponse_SpeechEventType_END_OF_SINGLE_UTTERANCE) {
+      // we only get this when we have requested it, and recognition stops after we get this
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got end_of_utterance\n") ;
+      cb->got_end_of_utterance = 1;
+      cb->responseHandler(session, "end_of_utterance", cb->bugname);
+      if (cb->wants_single_utterance) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: sending writesDone because we want only a single utterance\n") ;
+        streamer->writesDone();
+      }
+    }
+    else if (speech_event_type == StreamingRecognizeResponse_SpeechEventType_SPEECH_ACTIVITY_BEGIN) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got SPEECH_ACTIVITY_BEGIN\n") ;
+    }
+    else if (speech_event_type == StreamingRecognizeResponse_SpeechEventType_SPEECH_ACTIVITY_END) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got SPEECH_ACTIVITY_END\n") ;
+    }
+    switch_core_session_rwunlock(session);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: got %d responses\n", response.results_size());
+  }
+
+  {
+    switch_core_session_t* session = switch_core_session_locate(cb->sessionId);
+    if (session) {
+      grpc::Status status = streamer->finish();
+      // TODO: This works on the same principle as that used in the v1 equivalent, in that we search for the textual
+      // error message to determine whether the cause of the problem is the expiration of the session.
+      // It would be better if we could find a more reliable way of detecting this.
+      if (10 == status.error_code()) {
+        if (std::string::npos != status.error_message().find("Max duration of 5 minutes reached")) {
+          cb->responseHandler(session, "max_duration_exceeded", cb->bugname);
+        }
+        else {
+          cb->responseHandler(session, "no_audio", cb->bugname);
+        }
+      }
+      else if (status.error_code() != 0) {
+        cJSON* json = cJSON_CreateObject();
+        cJSON_AddStringToObject(json, "type", "error");
+        cJSON_AddStringToObject(json, "error_cause", "stream_close");
+        cJSON_AddItemToObject(json, "error_code", cJSON_CreateNumber(status.error_code()));
+        cJSON_AddStringToObject(json, "error_message", status.error_message().c_str());
+        char* jsonString = cJSON_PrintUnformatted(json);
+        cb->responseHandler(session, jsonString, cb->bugname);
+        free(jsonString);
+        cJSON_Delete(json);
+      }
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "grpc_read_thread: finish() status %s (%d)\n", status.error_message().c_str(), status.error_code()) ;
+      switch_core_session_rwunlock(session);
+    }
+  }
+  return nullptr;
+}
+
+template <>
+bool GStreamer<StreamingRecognizeRequest, StreamingRecognizeResponse, Speech::Stub>::write(void* data, uint32_t datalen) {
+	if (!m_connected) {
+		if (datalen % CHUNKSIZE == 0) {
+			m_audioBuffer.add(data, datalen);
+		}
+		return true;
+	}
+    m_request.clear_streaming_config();
+	m_request.set_audio(data, datalen);
+	bool ok = m_streamer->Write(m_request);
+	return ok;
+}
+
+extern "C" {
+
+    switch_status_t google_speech_session_cleanup_v2(switch_core_session_t *session, int channelIsClosing, switch_media_bug_t *bug) {
+      return google_speech_session_cleanup<GStreamer_V2>(session, channelIsClosing, bug);
+    }
+
+    switch_bool_t google_speech_frame_v2(switch_media_bug_t *bug, void* user_data) {
+      return google_speech_frame<GStreamer_V2>(bug, user_data);
+    }
+
+    switch_status_t google_speech_session_init_v2(switch_core_session_t *session, responseHandler_t responseHandler,
+		  uint32_t to_rate, uint32_t samples_per_second, uint32_t channels, char* lang, int interim, char *bugname, int single_utterance,
+		  int separate_recognition, int max_alternatives, int profanity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, 
+		  const char* hints, char* play_file, void **ppUserData) {
+      return google_speech_session_init<GStreamer_V2>(session, responseHandler, grpc_read_thread, to_rate, samples_per_second, channels,
+        lang, interim, bugname, single_utterance, separate_recognition, max_alternatives, profanity_filter,
+        word_time_offset, punctuation, model, enhanced, hints, play_file, ppUserData);
+    }
+
+}

mod_google_transcribe/gstreamer.h

@@ -0,0 +1,147 @@
+#include <cstdlib>
+#include <algorithm>
+#include <future>
+
+#include <switch.h>
+#include <switch_json.h>
+#include <grpc++/grpc++.h>
+#include <grpcpp/impl/codegen/sync_stream.h>
+
+#include "mod_google_transcribe.h"
+#include "simple_buffer.h"
+
+#define CHUNKSIZE (320)
+
+namespace {
+  int case_insensitive_match(std::string s1, std::string s2) {
+   std::transform(s1.begin(), s1.end(), s1.begin(), ::tolower);
+   std::transform(s2.begin(), s2.end(), s2.begin(), ::tolower);
+   if(s1.compare(s2) == 0)
+      return 1; //The strings are same
+   return 0; //not matched
+  }
+}
+
+template <typename Request, typename Response, typename Stub>
+class GStreamer {
+public:
+	GStreamer(
+    switch_core_session_t *session, 
+    uint32_t channels, 
+    char* lang, 
+    int interim, 
+    uint32_t config_sample_rate,
+		uint32_t samples_per_second,
+    int single_utterance, 
+    int separate_recognition,
+		int max_alternatives, 
+    int profanity_filter, 
+    int word_time_offset, 
+    int punctuation, 
+    const char* model, 
+    int enhanced, 
+	const char* hints);
+
+	~GStreamer() {
+		//switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(m_session), SWITCH_LOG_INFO, "GStreamer::~GStreamer - deleting channel and stub: %p\n", (void*)this);
+	}
+
+	bool write(void* data, uint32_t datalen);
+
+	void connect() {
+		assert(!m_connected);
+		// Begin a stream.
+		m_streamer = m_stub->StreamingRecognize(&m_context);
+		m_connected = true;
+
+		// read thread is waiting on this
+		m_promise.set_value();
+
+		// Write the first request, containing the config only.
+		m_streamer->Write(m_request);
+
+		// send any buffered audio
+		int nFrames = m_audioBuffer.getNumItems();
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer %p got stream ready, %d buffered frames\n", this, nFrames);	
+		if (nFrames) {
+			char *p;
+			do {
+				p = m_audioBuffer.getNextChunk();
+				if (p) {
+					write(p, CHUNKSIZE);
+				}
+			} while (p);
+		}
+	}
+
+	uint32_t nextMessageSize(void) {
+		uint32_t size = 0;
+		m_streamer->NextMessageSize(&size);
+		return size;
+	}
+
+	bool read(Response* response) {
+		return m_streamer->Read(response);
+	}
+
+	grpc::Status finish() {
+		return m_streamer->Finish();
+	}
+
+	void writesDone() {
+		// grpc crashes if we call this twice on a stream
+		if (!m_connected) {
+			cancelConnect();
+		}
+		else if (!m_writesDone) {
+			m_streamer->WritesDone();
+			m_writesDone = true;
+		}
+	}
+
+	bool waitForConnect() {
+		std::shared_future<void> sf(m_promise.get_future());
+		sf.wait();
+		return m_connected;
+	}
+
+	void cancelConnect() {
+		assert(!m_connected);
+		m_promise.set_value();
+	}
+
+	bool isConnected() {
+		return m_connected;
+	}
+
+private:
+	std::shared_ptr<grpc::Channel> create_grpc_channel(switch_channel_t *channel) {
+	    const char* google_uri;
+		if (!(google_uri = switch_channel_get_variable(channel, "GOOGLE_SPEECH_TO_TEXT_URI"))) {
+			google_uri = "speech.googleapis.com";
+		}
+
+	    const char* var;
+		if (var = switch_channel_get_variable(channel, "GOOGLE_APPLICATION_CREDENTIALS")) {
+			auto channelCreds = grpc::SslCredentials(grpc::SslCredentialsOptions());
+			auto callCreds = grpc::ServiceAccountJWTAccessCredentials(var);
+			auto creds = grpc::CompositeChannelCredentials(channelCreds, callCreds);
+			return grpc::CreateChannel(google_uri, creds);
+		}
+		else {
+			auto creds = grpc::GoogleDefaultCredentials();
+			return grpc::CreateChannel(google_uri, creds);
+		}
+	}
+
+	switch_core_session_t* m_session;
+	grpc::ClientContext m_context;
+	std::shared_ptr<grpc::Channel> m_channel;
+	std::unique_ptr<Stub> 	m_stub;
+	std::unique_ptr< grpc::ClientReaderWriterInterface<Request, Response> > m_streamer;
+	Request m_request;
+	bool m_writesDone;
+	bool m_connected;
+	std::promise<void> m_promise;
+	SimpleBuffer m_audioBuffer;
+};

mod_google_transcribe/mod_google_transcribe.c

@@ -10,6 +10,13 @@
 
 static const uint32_t DEFAULT_SAMPLE_RATE = 8000;
 
+/* Callback Type Definitions */
+typedef switch_status_t (*speech_cleanup_callback_t) (switch_core_session_t *, int, switch_media_bug_t *);
+typedef switch_bool_t (*speech_frame_callback_t) (switch_media_bug_t *, void *);
+typedef switch_status_t (*speech_init_callback_t) (switch_core_session_t *, responseHandler_t,
+		  uint32_t, uint32_t, uint32_t, char *, int, char *, int, int, int, int, int, int, const char *,
+		  int, const char *, char *, void **);
+
 /* Prototypes */
 SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_transcribe_shutdown);
 SWITCH_MODULE_RUNTIME_FUNCTION(mod_transcribe_runtime);
@@ -17,7 +24,42 @@ SWITCH_MODULE_LOAD_FUNCTION(mod_transcribe_load);
 
 SWITCH_MODULE_DEFINITION(mod_google_transcribe, mod_transcribe_load, mod_transcribe_shutdown, NULL);
 
-static switch_status_t do_stop(switch_core_session_t *session, char* bugname);
+static switch_bool_t capture_callback_v1(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type);
+static switch_bool_t capture_callback_v2(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type);
+static switch_status_t do_stop(switch_core_session_t *session, char* bugname, speech_cleanup_callback_t cleanup_callback);
+
+static switch_media_bug_callback_t get_bug_callback_from_version(GoogleCloudServiceVersion version) {
+	switch (version) {
+		case GoogleCloudServiceVersion_v1:
+			return capture_callback_v1;
+		case GoogleCloudServiceVersion_v2:
+			return capture_callback_v2;
+		default:
+			return capture_callback_v1;
+	}
+}
+
+static speech_cleanup_callback_t get_cleanup_callback_from_version(GoogleCloudServiceVersion version) {
+	switch (version) {
+		case GoogleCloudServiceVersion_v1:
+			return google_speech_session_cleanup_v1;
+		case GoogleCloudServiceVersion_v2:
+			return google_speech_session_cleanup_v2;
+		default:
+			return google_speech_session_cleanup_v1;
+	}
+}
+
+static speech_init_callback_t get_init_callback_from_version(GoogleCloudServiceVersion version) {
+	switch (version) {
+		case GoogleCloudServiceVersion_v1:
+			return google_speech_session_init_v1;
+		case GoogleCloudServiceVersion_v2:
+			return google_speech_session_init_v2;
+		default:
+			return google_speech_session_init_v1;
+	}
+}
 
 
 static void responseHandler(switch_core_session_t* session, const char * json, const char* bugname) {
@@ -34,6 +76,16 @@ static void responseHandler(switch_core_session_t* session, const char * json, c
 		switch_channel_event_set_data(channel, event);
 		switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-vendor", "google");
 	}
+	else if (0 == strcmp("start_of_speech", json)) {
+		switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_START_OF_SPEECH);
+		switch_channel_event_set_data(channel, event);
+		switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-vendor", "google");
+	}
+	else if (0 == strcmp("end_of_speech", json)) {
+		switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_END_OF_SPEECH);
+		switch_channel_event_set_data(channel, event);
+		switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-vendor", "google");
+	}
 	else if (0 == strcmp("end_of_transcript", json)) {
 		switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_END_OF_TRANSCRIPT);
 		switch_channel_event_set_data(channel, event);
@@ -69,19 +121,19 @@ static void responseHandler(switch_core_session_t* session, const char * json, c
 		switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-vendor", "google");
 	}
 	else {
-    int error = 0;
-    cJSON* jMessage = cJSON_Parse(json);
-    if (jMessage) {
-      const char* type = cJSON_GetStringValue(cJSON_GetObjectItem(jMessage, "type"));
-      if (type && 0 == strcmp(type, "error")) {
-        error = 1;
-    		switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_ERROR);
-      }
-      cJSON_Delete(jMessage);
-    }
-    if (!error) {
-    		switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_RESULTS);
-    }
+		int error = 0;
+		cJSON* jMessage = cJSON_Parse(json);
+		if (jMessage) {
+			const char* type = cJSON_GetStringValue(cJSON_GetObjectItem(jMessage, "type"));
+			if (type && 0 == strcmp(type, "error")) {
+				error = 1;
+				switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_ERROR);
+			}
+			cJSON_Delete(jMessage);
+		}
+		if (!error) {
+			switch_event_create_subclass(&event, SWITCH_EVENT_CUSTOM, TRANSCRIBE_EVENT_RESULTS);
+		}
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "%s json payload: %s.\n", bugname ? bugname : "google_transcribe", json);
 		switch_channel_event_set_data(channel, event);
 		switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "transcription-vendor", "google");
@@ -91,29 +143,29 @@ static void responseHandler(switch_core_session_t* session, const char * json, c
 	switch_event_fire(&event);
 }
 
-static switch_bool_t capture_callback(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type)
+static switch_bool_t capture_callback(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type,
+	speech_frame_callback_t frame_callback, speech_cleanup_callback_t cleanup_callback)
 {
 	switch_core_session_t *session = switch_core_media_bug_get_session(bug);
 	struct cap_cb* cb = (struct cap_cb*) switch_core_media_bug_get_user_data(bug);
 
 	switch (type) {
 	case SWITCH_ABC_TYPE_INIT:
-			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Got SWITCH_ABC_TYPE_INIT.\n");
-			responseHandler(session, "start_of_transcript", cb->bugname);
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Got SWITCH_ABC_TYPE_INIT.\n");
+		responseHandler(session, "start_of_transcript", cb->bugname);
 		break;
 
 	case SWITCH_ABC_TYPE_CLOSE:
 		{
 			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Got SWITCH_ABC_TYPE_CLOSE, calling google_speech_session_cleanup.\n");
 			responseHandler(session, "end_of_transcript", cb->bugname);
-			google_speech_session_cleanup(session, 1, bug);
+			cleanup_callback(session, 1, bug);
 			switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Finished SWITCH_ABC_TYPE_CLOSE.\n");
 		}
 		break;
 	
 	case SWITCH_ABC_TYPE_READ:
-
-		return google_speech_frame(bug, user_data);
+		return frame_callback(bug, user_data);
 		break;
 
 	case SWITCH_ABC_TYPE_WRITE:
@@ -124,19 +176,27 @@ static switch_bool_t capture_callback(switch_media_bug_t *bug, void *user_data,
 	return SWITCH_TRUE;
 }
 
+static switch_bool_t capture_callback_v1(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type) {
+	return capture_callback(bug, user_data, type, google_speech_frame_v1, google_speech_session_cleanup_v1);
+}
+
+static switch_bool_t capture_callback_v2(switch_media_bug_t *bug, void *user_data, switch_abc_type_t type) {
+	return capture_callback(bug, user_data, type, google_speech_frame_v2, google_speech_session_cleanup_v2);
+}
+
 static switch_status_t transcribe_input_callback(switch_core_session_t *session, void *input, switch_input_type_t input_type, void *data, unsigned int len){
 	if (input_type == SWITCH_INPUT_TYPE_EVENT) {
-			switch_event_t *event;
-			event = (switch_event_t *)input;
-			if (event->event_id == SWITCH_EVENT_DETECTED_SPEECH) {
-				return SWITCH_STATUS_BREAK;
-			}
+		switch_event_t *event;
+		event = (switch_event_t *)input;
+		if (event->event_id == SWITCH_EVENT_DETECTED_SPEECH) {
+			return SWITCH_STATUS_BREAK;
+		}
 	}
 	
 	return SWITCH_STATUS_SUCCESS;
 }
 
-static switch_status_t do_stop(switch_core_session_t *session, char *bugname)
+static switch_status_t do_stop(switch_core_session_t *session, char *bugname, speech_cleanup_callback_t cleanup_callback)
 {
 	switch_status_t status = SWITCH_STATUS_SUCCESS;
 	switch_channel_t *channel = switch_core_session_get_channel(session);
@@ -144,7 +204,7 @@ static switch_status_t do_stop(switch_core_session_t *session, char *bugname)
 
 	if (bug) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "Received user command command, calling google_speech_session_cleanup (possibly to stop prev transcribe)\n");
-		status = google_speech_session_cleanup(session, 0, bug);
+		status = cleanup_callback(session, 0, bug);
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "stopped transcription.\n");
 	}
 
@@ -153,7 +213,8 @@ static switch_status_t do_stop(switch_core_session_t *session, char *bugname)
 
 static switch_status_t start_capture2(switch_core_session_t *session, switch_media_bug_flag_t flags, 
   uint32_t sample_rate, char* lang, int interim, int single_utterance, int separate_recognition, int max_alternatives,
-  int profinity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, const char* hints, char* play_file)
+  int profanity_filter, int word_time_offset, int punctuation, const char* model, int enhanced, const char* hints,
+  char* play_file, GoogleCloudServiceVersion version)
 {
 	switch_channel_t *channel = switch_core_session_get_channel(session);
 	switch_media_bug_t *bug;
@@ -165,7 +226,7 @@ static switch_status_t start_capture2(switch_core_session_t *session, switch_med
 
 	if (switch_channel_get_private(channel, MY_BUG_NAME)) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing bug from previous transcribe\n");
-		do_stop(session, MY_BUG_NAME);
+		do_stop(session, MY_BUG_NAME, get_cleanup_callback_from_version(version));
 	}
 
 	switch_core_session_get_read_impl(session, &read_impl);
@@ -175,13 +236,15 @@ static switch_status_t start_capture2(switch_core_session_t *session, switch_med
 	}
 
 	samples_per_second = !strcasecmp(read_impl.iananame, "g722") ? read_impl.actual_samples_per_second : read_impl.samples_per_second;
+	status = get_init_callback_from_version(version)(session, responseHandler, sample_rate, samples_per_second, flags & SMBF_STEREO ? 2 : 1, lang, interim, MY_BUG_NAME, single_utterance,
+			separate_recognition, max_alternatives, profanity_filter, word_time_offset, punctuation, model, enhanced, hints, play_file, &pUserData);
 
-	if (SWITCH_STATUS_FALSE == google_speech_session_init(session, responseHandler, sample_rate, samples_per_second, flags & SMBF_STEREO ? 2 : 1, lang, interim, MY_BUG_NAME, single_utterance,
-	 separate_recognition, max_alternatives, profinity_filter, word_time_offset, punctuation, model, enhanced, hints, play_file, &pUserData)) {
+	if (SWITCH_STATUS_FALSE == status) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing google speech session.\n");
 		return SWITCH_STATUS_FALSE;
 	}
-	if ((status = switch_core_media_bug_add(session, "google_transcribe", NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
+
+	if ((status = switch_core_media_bug_add(session, "google_transcribe", NULL, get_bug_callback_from_version(version), pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
 		return status;
 	}
 
@@ -197,22 +260,23 @@ static switch_status_t start_capture2(switch_core_session_t *session, switch_med
 }
 
 static switch_status_t start_capture(switch_core_session_t *session, switch_media_bug_flag_t flags, 
-  char* lang, int interim, char* bugname)
+  char* lang, int interim, char* bugname, GoogleCloudServiceVersion version)
 {
 	switch_channel_t *channel = switch_core_session_get_channel(session);
 	switch_media_bug_t *bug;
 	switch_status_t status;
 	switch_codec_implementation_t read_impl = { 0 };
 	void *pUserData;
+	uint32_t sample_rate = DEFAULT_SAMPLE_RATE;
 	uint32_t samples_per_second;
 	int single_utterance = 0, separate_recognition = 0, max_alternatives = 0, profanity_filter = 0, word_time_offset = 0, punctuation = 0, enhanced = 0;
 	const char* hints = NULL;
-  const char* model = NULL;
+  	const char* model = NULL;
 	const char* var;
-
+	
 	if (switch_channel_get_private(channel, bugname)) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing bug from previous transcribe\n");
-		do_stop(session, bugname);
+		do_stop(session, bugname, get_cleanup_callback_from_version(version));
 	}
 
 	if (switch_true(switch_channel_get_variable(channel, "GOOGLE_SPEECH_SINGLE_UTTERANCE"))) {
@@ -229,6 +293,11 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
      	max_alternatives = atoi(var);
     }
 
+	// sample rate
+	if ((var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_SAMPLE_RATE"))) {
+     	sample_rate = atoi(var);
+    }
+
 	// profanity filter
 	if (switch_true(switch_channel_get_variable(channel, "GOOGLE_SPEECH_PROFANITY_FILTER"))) {
       profanity_filter = 1;
@@ -266,14 +335,16 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	}
 
 	samples_per_second = !strcasecmp(read_impl.iananame, "g722") ? read_impl.actual_samples_per_second : read_impl.samples_per_second;
+	status = SWITCH_STATUS_FALSE;
+	status = get_init_callback_from_version(version)(session, responseHandler, sample_rate, samples_per_second, flags & SMBF_STEREO ? 2 : 1, lang, interim, bugname, single_utterance,
+			separate_recognition, max_alternatives, profanity_filter, word_time_offset, punctuation, model, enhanced, hints, NULL, &pUserData);
 
-	if (SWITCH_STATUS_FALSE == google_speech_session_init(session, responseHandler, DEFAULT_SAMPLE_RATE, samples_per_second, flags & SMBF_STEREO ? 2 : 1, lang, interim, bugname, single_utterance,
-	 separate_recognition, max_alternatives, profanity_filter, word_time_offset, punctuation, model, enhanced, hints, NULL, &pUserData)) {
+	if (SWITCH_STATUS_FALSE == status) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error initializing google speech session.\n");
 		return SWITCH_STATUS_FALSE;
 	}
 
-	if ((status = switch_core_media_bug_add(session, bugname, NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
+	if ((status = switch_core_media_bug_add(session, bugname, NULL, get_bug_callback_from_version(version), pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
 		return status;
 	}
 
@@ -283,7 +354,7 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 }
 
 // #define TRANSCRIBE_API_SYNTAX "<uuid> [start|stop] [lang-code] [interim] [single-utterance](bool) [seperate-recognition](bool) [max-alternatives](int) [profinity-filter](bool) [word-time](bool) [punctuation](bool) [model](string) [enhanced](true) [hints](string without space) [play-file]"
-#define TRANSCRIBE2_API_SYNTAX "<uuid> [start|stop] [lang-code] [interim]  [single-utterance] [seperate-recognition] [max-alternatives] [profinity-filter] [word-time] [punctuation] [sample-rate] [model] [enhanced] [hints] [play-file]"
+#define TRANSCRIBE2_API_SYNTAX "<uuid> [start|stop] [lang-code] [interim]  [single-utterance] [separate-recognition] [max-alternatives] [profanity-filter] [word-time] [punctuation] [sample-rate] [model] [enhanced] [hints] [play-file]"
 SWITCH_STANDARD_API(transcribe2_function)
 {
 	char *mycmd = NULL, *argv[20] = { 0 };
@@ -295,6 +366,9 @@ SWITCH_STANDARD_API(transcribe2_function)
 	
 	switch_status_t status = SWITCH_STATUS_FALSE;
 	switch_media_bug_flag_t flags = SMBF_READ_STREAM /* | SMBF_WRITE_STREAM | SMBF_READ_PING */;
+	switch_channel_t *channel;
+	const char* var;
+	GoogleCloudServiceVersion version = GoogleCloudServiceVersion_v1;
 
 	if (!zstr(cmd) && (mycmd = strdup(cmd))) {
 		argc = switch_separate_string(mycmd, ' ', argv, (sizeof(argv) / sizeof(argv[0])));
@@ -311,16 +385,24 @@ SWITCH_STANDARD_API(transcribe2_function)
 		switch_core_session_t *lsession = NULL;
 
 		if ((lsession = switch_core_session_locate(argv[0]))) {
+			channel = switch_core_session_get_channel(lsession);
+			if ((var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_CLOUD_SERVICES_VERSION"))) {
+				if (!strcasecmp(var, "v1"))
+					version = GoogleCloudServiceVersion_v1;
+				else if (!strcasecmp(var, "v2"))
+					version = GoogleCloudServiceVersion_v2;
+			}
+
 			if (!strcasecmp(argv[1], "stop")) {
     		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "stop transcribing\n");
-				status = do_stop(lsession, MY_BUG_NAME);
+				status = do_stop(lsession, MY_BUG_NAME, get_cleanup_callback_from_version(version));
 			} else if (!strcasecmp(argv[1], "start")) {
-        char* lang = argv[2];
-        int interim = argc > 3 && !strcmp(argv[3], "true");
+        		char* lang = argv[2];
+        		int interim = argc > 3 && !strcmp(argv[3], "true");
 				int single_utterance =  !strcmp(argv[4], "true"); // single-utterance
-				int separate_recognition = !strcmp(argv[5], "true"); // sepreate-recognition
+				int separate_recognition = !strcmp(argv[5], "true"); // separate-recognition
 				int max_alternatives = atoi(argv[6]); // max-alternatives
-				int profinity_filter = !strcmp(argv[7], "true"); // profinity-filter
+				int profanity_filter = !strcmp(argv[7], "true"); // profanity-filter
 				int word_time_offset = !strcmp(argv[8], "true"); // word-time
 				int punctuation      = !strcmp(argv[9], "true");  //punctuation
 				if (argc > 10) {
@@ -336,9 +418,9 @@ SWITCH_STANDARD_API(transcribe2_function)
 				if (argc > 14){
 					play_file = argv[14];
 				}
-    		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "start transcribing %s %s\n", lang, interim ? "interim": "complete");
+    			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "start transcribing %s %s\n", lang, interim ? "interim": "complete");
 				status = start_capture2(lsession, flags, sample_rate, lang, interim, single_utterance, separate_recognition,max_alternatives,
-				profinity_filter, word_time_offset, punctuation, model, enhanced, hints, play_file);
+				profanity_filter, word_time_offset, punctuation, model, enhanced, hints, play_file, version);
 			}
 			switch_core_session_rwunlock(lsession);
 		}
@@ -363,6 +445,9 @@ SWITCH_STANDARD_API(transcribe_function)
 	int argc = 0;
 	switch_status_t status = SWITCH_STATUS_FALSE;
 	switch_media_bug_flag_t flags = SMBF_READ_STREAM /* | SMBF_WRITE_STREAM | SMBF_READ_PING */;
+	switch_channel_t *channel;
+	const char* var;
+	GoogleCloudServiceVersion version = GoogleCloudServiceVersion_v1;
 
 	if (!zstr(cmd) && (mycmd = strdup(cmd))) {
 		argc = switch_separate_string(mycmd, ' ', argv, (sizeof(argv) / sizeof(argv[0])));
@@ -379,20 +464,28 @@ SWITCH_STANDARD_API(transcribe_function)
 		switch_core_session_t *lsession = NULL;
 
 		if ((lsession = switch_core_session_locate(argv[0]))) {
+			channel = switch_core_session_get_channel(lsession);
+			if ((var = switch_channel_get_variable(channel, "GOOGLE_SPEECH_CLOUD_SERVICES_VERSION"))) {
+				if (!strcasecmp(var, "v1"))
+					version = GoogleCloudServiceVersion_v1;
+				else if (!strcasecmp(var, "v2"))
+					version = GoogleCloudServiceVersion_v2;
+			}
+
 			if (!strcasecmp(argv[1], "stop")) {
 				char *bugname = argc > 2 ? argv[2] : MY_BUG_NAME;
-    		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "stop transcribing\n");
-				status = do_stop(lsession, bugname);
+	    		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "stop transcribing\n");
+				status = do_stop(lsession, bugname, get_cleanup_callback_from_version(version));
 			} else if (!strcasecmp(argv[1], "start")) {
-        char* lang = argv[2];
-        int interim = argc > 3 && !strcmp(argv[3], "interim");
+    		    char* lang = argv[2];
+        		int interim = argc > 3 && !strcmp(argv[3], "interim");
 				char *bugname = argc > 5 ? argv[5] : MY_BUG_NAME;
 				if (argc > 4 && !strcmp(argv[4], "stereo")) {
-          flags |= SMBF_WRITE_STREAM ;
-          flags |= SMBF_STEREO;
+          			flags |= SMBF_WRITE_STREAM ;
+          			flags |= SMBF_STEREO;
 				}
-    		switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s start transcribing %s %s\n", bugname, lang, interim ? "interim": "complete");
-				status = start_capture(lsession, flags, lang, interim, bugname);
+    			switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "%s start transcribing %s %s\n", bugname, lang, interim ? "interim": "complete");
+				status = start_capture(lsession, flags, lang, interim, bugname, version);
 			}
 			switch_core_session_rwunlock(lsession);
 		}
@@ -423,6 +516,14 @@ SWITCH_MODULE_LOAD_FUNCTION(mod_transcribe_load)
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't register subclass %s!\n", TRANSCRIBE_EVENT_END_OF_UTTERANCE);
 		return SWITCH_STATUS_TERM;
 	}
+	if (switch_event_reserve_subclass(TRANSCRIBE_EVENT_START_OF_SPEECH) != SWITCH_STATUS_SUCCESS) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't register subclass %s!\n", TRANSCRIBE_EVENT_START_OF_SPEECH);
+		return SWITCH_STATUS_TERM;
+	}
+	if (switch_event_reserve_subclass(TRANSCRIBE_EVENT_END_OF_SPEECH) != SWITCH_STATUS_SUCCESS) {
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't register subclass %s!\n", TRANSCRIBE_EVENT_END_OF_SPEECH);
+		return SWITCH_STATUS_TERM;
+	}
 	if (switch_event_reserve_subclass(TRANSCRIBE_EVENT_START_OF_TRANSCRIPT) != SWITCH_STATUS_SUCCESS) {
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Couldn't register subclass %s!\n", TRANSCRIBE_EVENT_START_OF_TRANSCRIPT);
 		return SWITCH_STATUS_TERM;
@@ -473,6 +574,8 @@ SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_transcribe_shutdown)
 	google_speech_cleanup();
 	switch_event_free_subclass(TRANSCRIBE_EVENT_RESULTS);
 	switch_event_free_subclass(TRANSCRIBE_EVENT_END_OF_UTTERANCE);
+	switch_event_free_subclass(TRANSCRIBE_EVENT_START_OF_SPEECH);
+	switch_event_free_subclass(TRANSCRIBE_EVENT_END_OF_SPEECH);
 	switch_event_free_subclass(TRANSCRIBE_EVENT_START_OF_TRANSCRIPT);
 	switch_event_free_subclass(TRANSCRIBE_EVENT_END_OF_TRANSCRIPT);
 	switch_event_free_subclass(TRANSCRIBE_EVENT_NO_AUDIO_DETECTED);
@@ -481,4 +584,3 @@ SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_transcribe_shutdown)
 	switch_event_free_subclass(TRANSCRIBE_EVENT_PLAY_INTERRUPT);
 	return SWITCH_STATUS_SUCCESS;
 }
-

mod_google_transcribe/mod_google_transcribe.h

@@ -11,6 +11,8 @@
 #define MY_BUG_NAME "google_transcribe"
 #define TRANSCRIBE_EVENT_RESULTS "google_transcribe::transcription"
 #define TRANSCRIBE_EVENT_END_OF_UTTERANCE "google_transcribe::end_of_utterance"
+#define TRANSCRIBE_EVENT_START_OF_SPEECH "google_transcribe::start_of_speech"
+#define TRANSCRIBE_EVENT_END_OF_SPEECH "google_transcribe::end_of_speech"
 #define TRANSCRIBE_EVENT_START_OF_TRANSCRIPT "google_transcribe::start_of_transcript"
 #define TRANSCRIBE_EVENT_END_OF_TRANSCRIPT "google_transcribe::end_of_transcript"
 #define TRANSCRIBE_EVENT_NO_AUDIO_DETECTED "google_transcribe::no_audio_detected"
@@ -37,6 +39,10 @@ struct cap_cb {
 #else
 /* per-channel data */
 typedef void (*responseHandler_t)(switch_core_session_t* session, const char* json, const char* bugname);
+typedef enum GoogleCloudServiceVersion {
+    GoogleCloudServiceVersion_v1,
+    GoogleCloudServiceVersion_v2
+} GoogleCloudServiceVersion;
 
 struct cap_cb {
 	switch_mutex_t *mutex;

mod_ibm_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_jambonz_transcribe/LICENSE

@@ -1,5 +0,0 @@
-Copyright 2023, Drachtio Communications Services, LLC
-
-This software is provided under a dual-licensing scheme, described here: https://github.com/jambonz/freeswitch-modules/blob/main/COPYING.
-
-Please refer to the above file and the links therein for the complete terms and conditions of each license. Your use of this software constitutes agreement to the terms of these licenses. If you have any questions regarding the licensing of this software, please consult a legal expert.

mod_jambonz_transcribe/audio_pipe.cpp

@@ -229,16 +229,6 @@ int AudioPipe::lws_callback(struct lws *wsi,
 }
 
 
-// static members
-static const lws_retry_bo_t retry = {
-    nullptr,   // retry_ms_table
-    0,         // retry_ms_table_count
-    0,         // conceal_count
-    UINT16_MAX,         // secs_since_valid_ping
-    UINT16_MAX,        // secs_since_valid_hangup
-    0          // jitter_percent
-};
-
 struct lws_context *AudioPipe::context = nullptr;
 std::thread AudioPipe::serviceThread;
 std::mutex AudioPipe::mutex_connects;
@@ -376,6 +366,28 @@ bool AudioPipe::lws_service_thread() {
     { NULL, NULL, 0, 0 }
   };
 
+  uint16_t secs_sinceq_valid_ping = UINT16_MAX;
+  uint16_t secs_since_valid_hangup = UINT16_MAX;
+
+  char* wsVar = std::getenv("WS_PING_INTERVAL");
+  if (wsVar != nullptr) {
+    secs_sinceq_valid_ping = std::atoi(wsVar);
+  }
+
+  wsVar = std::getenv("WS_NO_PONG_HANGUP_INTERVAL");
+  if (wsVar != nullptr) {
+    secs_since_valid_hangup = std::atoi(wsVar);
+  }
+
+  const lws_retry_bo_t retry = {
+    nullptr,                  // retry_ms_table
+    0,                        // retry_ms_table_count
+    0,                        // conceal_count
+    secs_sinceq_valid_ping,   // secs_sinceq_valid_ping
+    secs_since_valid_hangup,  // secs_since_valid_hangup
+    0                         // jitter_percent
+  };
+
   memset(&info, 0, sizeof info); 
   info.port = CONTEXT_PORT_NO_LISTEN; 
   info.options = LWS_SERVER_OPTION_DO_SSL_GLOBAL_INIT;

mod_jambonz_transcribe/jb_transcribe_glue.cpp

@@ -145,7 +145,7 @@ namespace {
     cJSON_AddStringToObject(json, "format", "raw");
     cJSON_AddStringToObject(json, "encoding", "LINEAR16");
     cJSON_AddBoolToObject(json, "interimResults", tech_pvt->interim);
-    cJSON_AddNumberToObject(json, "sampleRateHz", 8000);
+    cJSON_AddNumberToObject(json, "sampleRateHz", tech_pvt->sampling);
     if (var = switch_channel_get_variable(channel, "JAMBONZ_STT_OPTIONS")) {
       cJSON* jOptions = cJSON_Parse(var);
       if (jOptions)  {
@@ -353,7 +353,7 @@ extern "C" {
   }
 	
   switch_status_t jb_transcribe_session_init(switch_core_session_t *session, 
-    responseHandler_t responseHandler, uint32_t samples_per_second, uint32_t channels, 
+    responseHandler_t responseHandler, uint32_t samples_per_second, int desiredSampling, uint32_t channels, 
     char* lang, int interim, char* bugname, void **ppUserData)
   {    	
     int err;
@@ -365,7 +365,7 @@ extern "C" {
       return SWITCH_STATUS_FALSE;
     }
 
-    if (SWITCH_STATUS_SUCCESS != fork_data_init(tech_pvt, session, samples_per_second, 8000, channels, lang, interim, bugname, responseHandler)) {
+    if (SWITCH_STATUS_SUCCESS != fork_data_init(tech_pvt, session, samples_per_second, desiredSampling, channels, lang, interim, bugname, responseHandler)) {
       destroy_tech_pvt(tech_pvt);
       return SWITCH_STATUS_FALSE;
     }
@@ -403,6 +403,7 @@ extern "C" {
         }
         jambonz::AudioPipe *pAudioPipe = static_cast<jambonz::AudioPipe *>(tech_pvt->pAudioPipe);
         if (pAudioPipe) reaper(tech_pvt);
+        switch_mutex_unlock(tech_pvt->mutex);
         destroy_tech_pvt(tech_pvt);
         switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) jb_transcribe_session_stop, bug removed\n", id);
         return SWITCH_STATUS_SUCCESS;
@@ -410,6 +411,7 @@ extern "C" {
       else {
         switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "jb_transcribe_session_stop: race condition, previous close completed\n");
       }
+      switch_mutex_unlock(tech_pvt->mutex);
     }
     return SWITCH_STATUS_FALSE;
   }