fix: gain was being ignored in sayOnTrack (#97)

* fixes for resampling and handling odd-number byte stream

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_audio_fork/audio_pipe.cpp

@@ -65,13 +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);
@@ -81,13 +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, 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;
@@ -102,7 +106,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
           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;
@@ -110,7 +114,7 @@ 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) {
@@ -119,7 +123,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
         }
         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());
+          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;
@@ -137,60 +141,68 @@ int AudioPipe::lws_callback(struct lws *wsi,
         
         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 (lws_frame_is_binary(wsi)) {
+        if (ap->m_state == LWS_CLIENT_DISCONNECTING) {
-          if (ap->is_bidirectional_audio_stream()) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO,"AudioPipe::lws_service_thread race condition: got incoming message while closing the connection.\n");
-            ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::BINARY, NULL, (char *) in, len);
-          } else {
-            lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE received binary frame, discarding.\n");
-          }
           return 0;
         }
 
-        if (lws_is_first_fragment(wsi)) {
+        if (lws_frame_is_binary(wsi)) {
-          // allocate a buffer for the entire chunk of memory needed
+          if (len > 0 && ap->is_bidirectional_audio_stream()) {
-          assert(nullptr == ap->m_recv_buf);
+            ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::BINARY, NULL, (char *) in, len);
-          ap->m_recv_buf_len = len + lws_remaining_packet_payload(wsi);
+          } else if (len > 0) {
-          ap->m_recv_buf = (uint8_t*) malloc(ap->m_recv_buf_len);
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR,"AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_RECEIVE received unexpected binary frame, discarding.\n");
-          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");
           }
           else {
-            ap->m_recv_buf = (uint8_t*) realloc(ap->m_recv_buf, newlen);
+            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) {
-              ap->m_recv_buf_len = newlen;
-              ap->m_recv_buf_ptr = ap->m_recv_buf + write_offset;
-            }
           }
         }
-
+        else {
-        if (nullptr != ap->m_recv_buf) {
+          if (lws_is_first_fragment(wsi)) {
-          if (len > 0) {
+            // allocate a buffer for the entire chunk of memory needed
-            memcpy(ap->m_recv_buf_ptr, in, len);
+            assert(nullptr == ap->m_recv_buf);
-            ap->m_recv_buf_ptr += len;
+            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) {
+          size_t write_offset = ap->m_recv_buf_ptr - ap->m_recv_buf;
-              std::string msg((char *)ap->m_recv_buf, ap->m_recv_buf_ptr - ap->m_recv_buf);
+          size_t remaining_space = ap->m_recv_buf_len - write_offset;
-              ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::MESSAGE, msg.c_str(), NULL, len);
+          if (remaining_space < len) {
-              if (nullptr != ap->m_recv_buf) free(ap->m_recv_buf);
+            //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;
           }
         }
       }
@@ -200,13 +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;
@@ -245,7 +257,7 @@ 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;
@@ -374,7 +386,7 @@ 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);
@@ -384,7 +396,7 @@ 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);
@@ -423,11 +435,11 @@ 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;
   }
 
@@ -444,16 +456,16 @@ 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()) {
@@ -508,7 +520,7 @@ 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;
 }

mod_audio_fork/lws_glue.cpp

@@ -21,11 +21,12 @@
 
 #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 (8192)
+#define BUFFER_GROW_SIZE (16384)
 
 namespace {
   static const char *requestedBufferSecs = std::getenv("MOD_AUDIO_FORK_BUFFER_SECS");
@@ -38,53 +39,146 @@ namespace {
   static uint32_t playCount = 0;
 
   switch_status_t processIncomingBinary(private_t* tech_pvt, switch_core_session_t* session, const char* message, size_t dataLength) {
-    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->circularBuffer;
+    std::vector<uint8_t> data;
-    uint8_t* data = reinterpret_cast<uint8_t*>(const_cast<char*>(message));
-    uint16_t* data_uint16 = reinterpret_cast<uint16_t*>(data);
-    std::vector<uint16_t> pcm_data(data_uint16, data_uint16 + dataLength / sizeof(uint16_t));
 
+    // 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;
+    }
 
-    if (tech_pvt->bidirectional_audio_resampler) {
+    // Append the new incoming message
-      std::vector<int16_t> in(pcm_data.begin(), pcm_data.end());
+    data.insert(data.end(), message, message + dataLength);
 
-      std::vector<int16_t> out(dataLength);
+    // Check if the total data length is now odd
-      spx_uint32_t in_len = pcm_data.size();
+    if (data.size() % 2 != 0) {
-      spx_uint32_t out_len = out.size();
+        // Set aside the last byte
-      speex_resampler_process_interleaved_int(tech_pvt->bidirectional_audio_resampler, in.data(), &in_len, out.data(), &out_len);
+        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
+    }
 
-      if (out_len > out.size()) {
+    // Convert the data to 16-bit elements
-        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_CRIT, "Resampler output exceeded maximum buffer size!\n");
+    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);
+
+    // Ensure the prebuffer has enough capacity
+    if (cBuffer->capacity() - cBuffer->size() < numSamples) {
+        size_t newCapacity = cBuffer->size() + std::max(numSamples, (size_t)BUFFER_GROW_SIZE);
+        cBuffer->set_capacity(newCapacity);
+    }
+
+    // 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();
 
-      // Resize the pcm_data to match the output length from resampler, and then copy the resampled data into it.
+      // Put the leftover samples back in the prebuffer for the next time
-      pcm_data.resize(out_len);
+      if (!leftoverSamples.empty()) {
-      memcpy(pcm_data.data(), out.data(), out_len * sizeof(int16_t));
+          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_mutex_lock(tech_pvt->mutex);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Failed to get mutext (temp)\n");
-
-    // Resize the buffer if necessary
-    size_t bytesResampled = pcm_data.size() * sizeof(uint16_t);
-    if (cBuffer->capacity() - cBuffer->size() < bytesResampled / sizeof(uint16_t)) {
-      // If buffer exceeds some max size, you could return SWITCH_STATUS_FALSE to abort the transfer
-      // if (cBuffer->size() + std::max(bytesResampled / sizeof(uint16_t), (size_t)BUFFER_GROW_SIZE) > MAX_BUFFER_SIZE) return SWITCH_STATUS_FALSE;
-
-      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.begin(), pcm_data.end());
-
-    switch_mutex_unlock(tech_pvt->mutex);
 
     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") &&
         // playAudio is enabled and there is no bidirectional audio from stream is enabled.
@@ -171,6 +265,9 @@ 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("transcription")) {
         char* jsonString = cJSON_PrintUnformatted(jsonData);
@@ -290,10 +387,20 @@ namespace {
     tech_pvt->buffer_overrun_notified = 0;
     tech_pvt->audio_paused = 0;
     tech_pvt->graceful_shutdown = 0;
-    tech_pvt->circularBuffer = (void *) new CircularBuffer_t(8192);
+    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);
+
     strncpy(tech_pvt->bugname, bugname, MAX_BUG_LEN);
     if (metadata) strncpy(tech_pvt->initialMetadata, metadata, MAX_METADATA_LEN);
     
@@ -323,8 +430,8 @@ namespace {
     }
 
     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\n", tech_pvt->id, bidirectional_audio_sample_rate, sampling);
+      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(channels, bidirectional_audio_sample_rate, sampling, SWITCH_RESAMPLE_QUALITY, &err);
+      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;
@@ -350,10 +457,15 @@ namespace {
       switch_mutex_destroy(tech_pvt->mutex);
       tech_pvt->mutex = nullptr;
     }
-    if (tech_pvt->circularBuffer) {
+    if (tech_pvt->streamingPlayoutBuffer) {
-      CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->circularBuffer;
+      CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
       delete cBuffer;
-      tech_pvt->circularBuffer = nullptr;
+      tech_pvt->streamingPlayoutBuffer = nullptr;
+    }
+    if (tech_pvt->streamingPreBuffer) {
+      CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPreBuffer;
+      delete cBuffer;
+      tech_pvt->streamingPreBuffer = nullptr;
     }
   }
 
@@ -453,8 +565,8 @@ 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 ;
+    //int logs = LLL_ERR | LLL_WARN | LLL_NOTICE | LLL_INFO | LLL_PARSER | LLL_HEADER | LLL_EXT | LLL_CLIENT  | LLL_LATENCY | LLL_DEBUG ;
-     //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;
@@ -706,28 +818,39 @@ extern "C" {
   }
 
   switch_bool_t dub_speech_frame(switch_media_bug_t *bug, private_t* tech_pvt) {
-    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->circularBuffer;
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
     if (switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
-      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;
+      // if flag was set to clear the buffer, do so and clear the flag
-      rframe->datalen = rframe->samples * sizeof(int16_t);
+      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); 
-      int16_t data[SWITCH_RECOMMENDED_BUFFER_SIZE];
+        cBuffer->clear();
-      memset(data, 0, sizeof(data));
+        tech_pvt->clear_bidirectional_audio_buffer = false;
-
-      int samplesToCopy = std::min(static_cast<int>(cBuffer->size()), static_cast<int>(rframe->samples));
-
-      std::copy_n(cBuffer->begin(), samplesToCopy, data);
-      cBuffer->erase(cBuffer->begin(), cBuffer->begin() + samplesToCopy);
-
-      if (samplesToCopy > 0) {
-        vector_add(fp, data, rframe->samples);
       }
-      vector_normalize(fp, rframe->samples);
+      else {
+        switch_frame_t* rframe = switch_core_media_bug_get_write_replace_frame(bug);
+        int16_t *fp = reinterpret_cast<int16_t*>(rframe->data);
 
-      switch_core_media_bug_set_write_replace_frame(bug, rframe);
+        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); 
+
+        std::copy_n(cBuffer->begin(), samplesToCopy, data);
+        cBuffer->erase(cBuffer->begin(), cBuffer->begin() + samplesToCopy);
+
+        if (samplesToCopy > 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(tech_pvt->mutex);
     }
     return SWITCH_TRUE;
@@ -742,7 +865,7 @@ extern "C" {
     }
     private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
 
-    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->circularBuffer;
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) tech_pvt->streamingPlayoutBuffer;
 
     if (switch_mutex_trylock(tech_pvt->mutex) == SWITCH_STATUS_SUCCESS) {
       if (cBuffer != nullptr) {

mod_audio_fork/mod_audio_fork.h

@@ -52,11 +52,19 @@ struct private_data {
   int audio_paused:1;
   int graceful_shutdown:1;
   char initialMetadata[8192];
-  void *circularBuffer;
+
+  // 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_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, 
@@ -141,7 +142,13 @@ public:
 			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
 		/*
@@ -246,6 +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);
 	}
 
@@ -253,6 +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;
@@ -304,7 +318,7 @@ public:
 
 	void finish() {
 		if (m_finished) return;
-		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "GStreamer::finish - calling  StopContinuousRecognitionAsync (%p)\n", this);
+		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "GStreamer::finish - calling  StopContinuousRecognitionAsync (%p)\n", this);
 		m_finished = true;
 		m_recognizer->StopContinuousRecognitionAsync().get();
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "GStreamer::finish - recognition has completed (%p)\n", this);
@@ -318,13 +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;
@@ -332,6 +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) {
@@ -388,16 +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);
@@ -418,15 +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);
@@ -468,23 +589,10 @@ extern "C" {
 					switch_channel_get_name(channel), voice_ms, mode);
 			}
 		}
-
+		if (!cb->vad) streamer->connect();
-		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;
-	
 	done:
+		*ppUserData = cb;
+		cb->is_keep_alive = 0;
 		return status;
 	}
 
@@ -496,6 +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");
@@ -507,6 +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;
@@ -524,7 +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;
-
+	int use_single_connection = switch_true(getenv("AZURE_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);
 	}
@@ -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) {
+	if (!bug || !use_single_connection) {
-		return status;
+		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/azure_glue.cpp

@@ -15,11 +15,13 @@ 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 \n");
+      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();
@@ -43,6 +45,8 @@ static void start_synthesis(std::shared_ptr<SpeechSynthesizer> speechSynthesizer
         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" {
@@ -87,14 +91,7 @@ extern "C" {
 
   switch_status_t azure_speech_feed_tts(azure_t* a, char* text, switch_speech_flag_t *flags) {
     const int MAX_CHARS = 20;
-    char tempText[MAX_CHARS + 4]; // +4 for the ellipsis and null terminator
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts %s\n", text);
-
-    if (strlen(text) > MAX_CHARS) {
-        strncpy(tempText, text, MAX_CHARS);
-        strcpy(tempText + MAX_CHARS, "...");
-    } else {
-        strcpy(tempText, text);
-    }
 
     /* open cache file */
     if (a->cache_audio && fullDirPath.length() > 0) {
@@ -168,8 +165,14 @@ extern "C" {
 			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);
+      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");
@@ -266,7 +269,8 @@ extern "C" {
         }
       };
 
-      std::thread(start_synthesis, speechSynthesizer, text, a).detach();
+      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());

mod_dub/mod_dub.c

@@ -159,14 +159,16 @@ static switch_status_t dub_silence_track(switch_core_session_t *session, char* t
   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 = silence_dub_track(cb, trackName);
+    status = remove_dub_track(cb, trackName);
     if (status != SWITCH_STATUS_SUCCESS) {
-      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "dub_silence_track: error silencing track %s\n", trackName);
+      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");
@@ -269,7 +271,7 @@ SWITCH_STANDARD_API(dub_function)
         }
         else {
           char* text = argv[3];
-          int gain = argc > 4 ? atoi(argv[4]) : 0;
+          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);
         }

mod_elevenlabs_tts/elevenlabs_glue.cpp

@@ -809,6 +809,12 @@ 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);

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;
@@ -174,6 +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

@@ -17,6 +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;