bug: fix crashing race condition in mod_dub

* support azure stream

* delete trash file

* wip

* wip

* wip

* wip

* wip

* wip

* wip

* fire variable_tts_time_to_first_byte_ms

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_assemblyai_transcribe/aai_transcribe_glue.cpp

@@ -102,11 +102,12 @@ namespace {
     return path;
   }
 
-  static void eventCallback(const char* sessionId, assemblyai::AudioPipe::NotifyEvent_t event, const char* message, bool finished) {
+  static void eventCallback(const char* sessionId, const char* bugname, 
+    assemblyai::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, MY_BUG_NAME);
+      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) {
@@ -194,7 +195,8 @@ namespace {
     tech_pvt->channels = channels;
     tech_pvt->id = ++idxCallCount;
     tech_pvt->buffer_overrun_notified = 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, "ASSEMBLYAI_API_KEY");
@@ -204,7 +206,7 @@ namespace {
       return SWITCH_STATUS_FALSE;
     }
 
-    assemblyai::AudioPipe* ap = new assemblyai::AudioPipe(tech_pvt->sessionId, tech_pvt->host, tech_pvt->port, tech_pvt->path, 
+    assemblyai::AudioPipe* ap = new assemblyai::AudioPipe(tech_pvt->sessionId, bugname, tech_pvt->host, tech_pvt->port, tech_pvt->path, 
       buflen, read_impl.decoded_bytes_per_packet, apiKey, eventCallback);
     if (!ap) {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error allocating AudioPipe\n");
@@ -251,7 +253,8 @@ extern "C" {
   switch_status_t aai_transcribe_init() {
     switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_assemblyai_transcribe: audio buffer (in secs):    %d secs\n", nAudioBufferSecs);
  
-    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 ;
     
     assemblyai::AudioPipe::initialize(logs, lws_logger);
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AudioPipe::initialize completed\n");
@@ -306,7 +309,7 @@ extern "C" {
 
 	switch_status_t aai_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);
+    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_DEBUG, "aai_transcribe_session_stop: no bug - websocket conection already closed\n");
       return SWITCH_STATUS_FALSE;

mod_assemblyai_transcribe/audio_pipe.cpp

@@ -72,7 +72,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
         lwsl_err("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(), AudioPipe::CONNECT_FAIL, (char *) in, ap->isFinished());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_FAIL, (char *) in, ap->isFinished());
         }
         else {
           lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR unable to find wsi %p..\n", wsi); 
@@ -87,7 +87,7 @@ 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(), AudioPipe::CONNECT_SUCCESS, NULL,  ap->isFinished());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_SUCCESS, NULL,  ap->isFinished());
         }
         else {
           lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_ESTABLISHED %s unable to find wsi %p..\n", ap->m_uuid.c_str(), wsi); 
@@ -105,12 +105,12 @@ int AudioPipe::lws_callback(struct lws *wsi,
           // closed by us
 
           lwsl_debug("%s socket closed by us\n", ap->m_uuid.c_str());
-          ap->m_callback(ap->m_uuid.c_str(), AudioPipe::CONNECTION_CLOSED_GRACEFULLY, NULL,  ap->isFinished());
+          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
           lwsl_info("%s socket closed by far end\n", ap->m_uuid.c_str());
-          ap->m_callback(ap->m_uuid.c_str(), AudioPipe::CONNECTION_DROPPED, NULL,  ap->isFinished());
+          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();
@@ -172,7 +172,7 @@ 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(), AudioPipe::MESSAGE, msg.c_str(),  ap->isFinished());
+              ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), 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;
@@ -261,6 +261,7 @@ static const lws_retry_bo_t retry = {
 };
 
 struct lws_context *AudioPipe::context = nullptr;
+std::thread AudioPipe::serviceThread;
 std::string AudioPipe::protocolName;
 std::mutex AudioPipe::mutex_connects;
 std::mutex AudioPipe::mutex_disconnects;
@@ -429,26 +430,28 @@ bool AudioPipe::lws_service_thread() {
 
 void AudioPipe::initialize(int loglevel, log_emit_function logger) {
 
-  lws_set_log_level(loglevel, logger);
+  //lws_set_log_level(loglevel, logger);
 
   lwsl_notice("AudioPipe::initialize starting\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
-  std::thread t(&AudioPipe::lws_service_thread);
   stopFlag = false;
-  t.detach();
+  serviceThread = std::thread(&AudioPipe::lws_service_thread);
 }
 
 bool AudioPipe::deinitialize() {
   lwsl_notice("AudioPipe::deinitialize\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = true;
+    if (serviceThread.joinable()) {
+    serviceThread.join();
+  }
   return true;
 }
 
 // instance members
-AudioPipe::AudioPipe(const char* uuid, const char* host, unsigned int port, const char* path,
+AudioPipe::AudioPipe(const char* uuid, const char* bugname, const char* host, unsigned int port, const char* path,
   size_t bufLen, size_t minFreespace, const char* apiKey, notifyHandler_t callback) :
-  m_uuid(uuid), m_host(host), m_port(port), m_path(path), m_finished(false),
+  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_state(LWS_CLIENT_IDLE), m_wsi(nullptr), m_vhd(nullptr), m_apiKey(apiKey), m_callback(callback) {

mod_assemblyai_transcribe/audio_pipe.hpp

@@ -31,7 +31,7 @@ public:
     MESSAGE
   };
   typedef void (*log_emit_function)(int level, const char *line);
-  typedef void (*notifyHandler_t)(const char *sessionId, NotifyEvent_t event, const char* message, bool finished);
+  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;
@@ -44,7 +44,7 @@ public:
   static bool lws_service_thread();
 
   // constructor
-  AudioPipe(const char* uuid, const char* host, unsigned int port, const char* path, 
+  AudioPipe(const char* uuid, const char* bugname, const char* host, unsigned int port, const char* path, 
     size_t bufLen, size_t minFreespace, const char* apiKey, notifyHandler_t callback);
   ~AudioPipe();  
 
@@ -86,6 +86,7 @@ public:
   void operator=(const AudioPipe&) = delete;
 
 private:
+  static std::thread serviceThread;
 
   static int lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
   static struct lws_context *context;

mod_assemblyai_transcribe/mod_assemblyai_transcribe.c

@@ -74,7 +74,7 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	void *pUserData;
 	uint32_t samples_per_second;
 
-	if (switch_channel_get_private(channel, MY_BUG_NAME)) {
+	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);
 	}
@@ -94,7 +94,7 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	if ((status = switch_core_media_bug_add(session, "aai_transcribe", NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
 		return status;
 	}
-  switch_channel_set_private(channel, MY_BUG_NAME, bug);
+  switch_channel_set_private(channel, bugname, bug);
 	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for assemblyai transcribe\n");
 
 	return SWITCH_STATUS_SUCCESS;
@@ -105,7 +105,7 @@ 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, MY_BUG_NAME);
+	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.\n");

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/audio_pipe.cpp

@@ -7,6 +7,7 @@
 #define MAX_RECV_BUF_SIZE (65 * 1024 * 10)
 #define RECV_BUF_REALLOC_SIZE (8 * 1024)
 
+using namespace drachtio;
 
 namespace {
   static const char* basicAuthUser = std::getenv("MOD_AUDIO_FORK_HTTP_AUTH_USER");
@@ -267,6 +268,7 @@ static const lws_retry_bo_t retry = {
 };
 
 struct lws_context *AudioPipe::context = nullptr;
+std::thread AudioPipe::serviceThread;
 std::string AudioPipe::protocolName;
 std::mutex AudioPipe::mutex_connects;
 std::mutex AudioPipe::mutex_disconnects;
@@ -436,19 +438,21 @@ 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"); 
   std::lock_guard<std::mutex> lock(mapMutex);
-  std::thread t(&AudioPipe::lws_service_thread);
   stopFlag = false;
-  t.detach();
+  serviceThread = std::thread(&AudioPipe::lws_service_thread);
 }
 
 bool AudioPipe::deinitialize() {
   lwsl_notice("AudioPipe::deinitialize\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = true;
+  if (serviceThread.joinable()) {
+    serviceThread.join();
+  }
   return true;
 }
 

mod_audio_fork/audio_pipe.hpp

@@ -10,135 +10,140 @@
 
 #include <libwebsockets.h>
 
-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);
+namespace drachtio {
 
-  struct lws_per_vhost_data {
-    struct lws_context *context;
-    struct lws_vhost *vhost;
-    const struct lws_protocols *protocol;
+  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);
+
+    struct lws_per_vhost_data {
+      struct lws_context *context;
+      struct lws_vhost *vhost;
+      const struct lws_protocols *protocol;
+    };
+
+    static void initialize(const char* protocolName, int loglevel, log_emit_function logger);
+    static bool deinitialize();
+    static bool lws_service_thread();
+
+    // 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);
+    ~AudioPipe();  
+
+    LwsState_t getLwsState(void) { return m_state; }
+    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) {
+      m_audio_buffer_write_offset = 0;
+    }
+    void lockAudioBuffer(void) {
+      m_audio_mutex.lock();
+    }
+    void unlockAudioBuffer(void) ;
+    bool hasBasicAuth(void) {
+      return !m_username.empty() && !m_password.empty();
+    }
+
+    void getBasicAuth(std::string& username, std::string& password) {
+      username = m_username;
+      password = m_password;
+    }
+
+    void do_graceful_shutdown();
+    bool isGracefulShutdown(void) {
+      return m_gracefulShutdown;
+    }
+
+    void close() ;
+
+    // no default constructor or copying
+    AudioPipe() = delete;
+    AudioPipe(const AudioPipe&) = delete;
+    void operator=(const AudioPipe&) = delete;
+
+  private:
+    static std::thread serviceThread;
+
+    static int lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
+    static struct lws_context *context;
+    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 bool stopFlag;
+
+    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;
+    std::string m_bugname;
+    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_username;
+    std::string m_password;
+    bool m_gracefulShutdown;
   };
 
-  static void initialize(const char* protocolName, int loglevel, log_emit_function logger);
-  static bool deinitialize();
-  static bool lws_service_thread();
-
-  // 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);
-  ~AudioPipe();  
-
-  LwsState_t getLwsState(void) { return m_state; }
-  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) {
-    m_audio_buffer_write_offset = 0;
-  }
-  void lockAudioBuffer(void) {
-    m_audio_mutex.lock();
-  }
-  void unlockAudioBuffer(void) ;
-  bool hasBasicAuth(void) {
-    return !m_username.empty() && !m_password.empty();
-  }
-
-  void getBasicAuth(std::string& username, std::string& password) {
-    username = m_username;
-    password = m_password;
-  }
-
-  void do_graceful_shutdown();
-  bool isGracefulShutdown(void) {
-    return m_gracefulShutdown;
-  }
-
-  void close() ;
-
-  // no default constructor or copying
-  AudioPipe() = delete;
-  AudioPipe(const AudioPipe&) = delete;
-  void operator=(const AudioPipe&) = delete;
-
-private:
-
-  static int lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
-  static struct lws_context *context;
-  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 bool stopFlag;
-
-  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;
-  std::string m_bugname;
-  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_username;
-  std::string m_password;
-  bool m_gracefulShutdown;
-};
+} // namespace drachtio
 
 #endif

mod_audio_fork/lws_glue.cpp

@@ -156,7 +156,7 @@ namespace {
     }
   }
 
-  static void eventCallback(const char* sessionId, const char* bugname, AudioPipe::NotifyEvent_t event, const char* message) {
+  static void eventCallback(const char* sessionId, const char* bugname, drachtio::AudioPipe::NotifyEvent_t event, const char* message) {
     switch_core_session_t* session = switch_core_session_locate(sessionId);
     if (session) {
       switch_channel_t *channel = switch_core_session_get_channel(session);
@@ -165,16 +165,16 @@ namespace {
         private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
         if (tech_pvt) {
           switch (event) {
-            case AudioPipe::CONNECT_SUCCESS:
+            case drachtio::AudioPipe::CONNECT_SUCCESS:
               switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "connection successful\n");
               tech_pvt->responseHandler(session, EVENT_CONNECT_SUCCESS, NULL);
               if (strlen(tech_pvt->initialMetadata) > 0) {
                 switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "sending initial metadata %s\n", tech_pvt->initialMetadata);
-                AudioPipe *pAudioPipe = static_cast<AudioPipe *>(tech_pvt->pAudioPipe);
+                drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe *>(tech_pvt->pAudioPipe);
                 pAudioPipe->bufferForSending(tech_pvt->initialMetadata);
               }
             break;
-            case AudioPipe::CONNECT_FAIL:
+            case drachtio::AudioPipe::CONNECT_FAIL:
             {
               // first thing: we can no longer access the AudioPipe
               std::stringstream json;
@@ -184,18 +184,18 @@ namespace {
               switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_NOTICE, "connection failed: %s\n", message);
             }
             break;
-            case AudioPipe::CONNECTION_DROPPED:
+            case drachtio::AudioPipe::CONNECTION_DROPPED:
               // first thing: we can no longer access the AudioPipe
               tech_pvt->pAudioPipe = nullptr;
               tech_pvt->responseHandler(session, EVENT_DISCONNECT, NULL);
               switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_NOTICE, "connection dropped from far end\n");
             break;
-            case AudioPipe::CONNECTION_CLOSED_GRACEFULLY:
+            case drachtio::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 AudioPipe::MESSAGE:
+            case drachtio::AudioPipe::MESSAGE:
               processIncomingMessage(tech_pvt, session, message);
             break;
           }
@@ -239,7 +239,7 @@ namespace {
     
     size_t buflen = LWS_PRE + (FRAME_SIZE_8000 * desiredSampling / 8000 * channels * 1000 / RTP_PACKETIZATION_PERIOD * nAudioBufferSecs);
 
-    AudioPipe* ap = new AudioPipe(tech_pvt->sessionId, host, port, path, sslFlags, 
+    drachtio::AudioPipe* ap = new drachtio::AudioPipe(tech_pvt->sessionId, host, port, path, sslFlags, 
       buflen, read_impl.decoded_bytes_per_packet, username, password, bugname, eventCallback);
     if (!ap) {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error allocating AudioPipe\n");
@@ -374,13 +374,17 @@ extern "C" {
  
     int logs = LLL_ERR | LLL_WARN | LLL_NOTICE ;
      //LLL_INFO | LLL_PARSER | LLL_HEADER | LLL_EXT | LLL_CLIENT  | LLL_LATENCY | LLL_DEBUG ;
-    AudioPipe::initialize(mySubProtocolName, logs, lws_logger);
-   return SWITCH_STATUS_SUCCESS;
+    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;
   }
 
   switch_status_t fork_cleanup() {
     bool cleanup = false;
-    cleanup = AudioPipe::deinitialize();
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_audio_fork unloading..\n");
+
+    cleanup = drachtio::AudioPipe::deinitialize();
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_audio_fork unloaded status %d\n", cleanup);
     if (cleanup == true) {
         return SWITCH_STATUS_SUCCESS;
     }
@@ -420,7 +424,7 @@ extern "C" {
 
    switch_status_t fork_session_connect(void **ppUserData) {
     private_t *tech_pvt = static_cast<private_t *>(*ppUserData);
-    AudioPipe *pAudioPipe = static_cast<AudioPipe*>(tech_pvt->pAudioPipe);
+    drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe*>(tech_pvt->pAudioPipe);
     pAudioPipe->connect();
     return SWITCH_STATUS_SUCCESS;
   }
@@ -438,7 +442,7 @@ extern "C" {
     switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "(%u) fork_session_cleanup\n", id);
 
     if (!tech_pvt) return SWITCH_STATUS_FALSE;
-    AudioPipe *pAudioPipe = static_cast<AudioPipe *>(tech_pvt->pAudioPipe);
+    drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe *>(tech_pvt->pAudioPipe);
       
     switch_mutex_lock(tech_pvt->mutex);
 
@@ -481,7 +485,7 @@ extern "C" {
     private_t* tech_pvt = (private_t*) switch_core_media_bug_get_user_data(bug);
   
     if (!tech_pvt) return SWITCH_STATUS_FALSE;
-    AudioPipe *pAudioPipe = static_cast<AudioPipe *>(tech_pvt->pAudioPipe);
+    drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe *>(tech_pvt->pAudioPipe);
     if (pAudioPipe && text) pAudioPipe->bufferForSending(text);
 
     return SWITCH_STATUS_SUCCESS;
@@ -516,7 +520,7 @@ extern "C" {
 
     tech_pvt->graceful_shutdown = 1;
 
-    AudioPipe *pAudioPipe = static_cast<AudioPipe *>(tech_pvt->pAudioPipe);
+    drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe *>(tech_pvt->pAudioPipe);
     if (pAudioPipe) pAudioPipe->do_graceful_shutdown();
 
     return SWITCH_STATUS_SUCCESS;
@@ -535,8 +539,8 @@ extern "C" {
         switch_mutex_unlock(tech_pvt->mutex);
         return SWITCH_TRUE;
       }
-      AudioPipe *pAudioPipe = static_cast<AudioPipe *>(tech_pvt->pAudioPipe);
-      if (pAudioPipe->getLwsState() != AudioPipe::LWS_CLIENT_CONNECTED) {
+      drachtio::AudioPipe *pAudioPipe = static_cast<drachtio::AudioPipe *>(tech_pvt->pAudioPipe);
+      if (pAudioPipe->getLwsState() != drachtio::AudioPipe::LWS_CLIENT_CONNECTED) {
         switch_mutex_unlock(tech_pvt->mutex);
         return SWITCH_TRUE;
       }

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_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

@@ -512,7 +512,18 @@ 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();

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

@@ -135,6 +135,13 @@ 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);
+		}
+
 		// recognition mode - readonly according to Azure docs: 
 		// https://docs.microsoft.com/en-us/javascript/api/microsoft-cognitiveservices-speech-sdk/propertyid?view=azure-node-latest
 		/*

mod_azure_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_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 `pkg-config --libs boost` -lstdc++

mod_azure_tts/azure_glue.cpp

@@ -0,0 +1,380 @@
+#include "mod_azure_tts.h"
+#include <switch.h>
+#include <speechapi_cxx.h>
+
+#include <boost/circular_buffer.hpp>
+
+#include <cstdlib>
+#include <string>
+#include <chrono>
+
+#define BUFFER_SIZE 8129
+
+typedef boost::circular_buffer<uint16_t> CircularBuffer_t;
+
+using namespace Microsoft::CognitiveServices::Speech;
+
+static std::string fullDirPath;
+
+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 = "/var/";
+    }
+    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";
+
+    // 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;
+    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 (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->session_id) {
+      int err;
+      switch_codec_implementation_t read_impl;
+      switch_core_session_t *psession = switch_core_session_locate(a->session_id);
+      switch_core_session_get_read_impl(psession, &read_impl);
+      uint32_t samples_per_second = !strcasecmp(read_impl.iananame, "g722") ? read_impl.actual_samples_per_second : read_impl.samples_per_second;
+      a->samples_rate = samples_per_second;
+      if (samples_per_second != 8000 /*Hz*/) {
+        a->resampler = speex_resampler_init(1, 8000, samples_per_second, 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);
+		}
+
+    auto speechSynthesizer = SpeechSynthesizer::FromConfig(speechConfig);
+
+    speechSynthesizer->SynthesisStarted += [a](const SpeechSynthesisEventArgs& e) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts SynthesisStarted\n");
+        a->response_code = 200;
+    };
+
+    speechSynthesizer->Synthesizing += [a](const SpeechSynthesisEventArgs& e) {
+      bool fireEvent = false;
+      CircularBuffer_t *cBuffer = (CircularBuffer_t *) a->circularBuffer;
+      std::vector<uint16_t> pcm_data;
+
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "Synthesizing: received data\n");
+
+      if (a->flushed) {
+        return;
+      }
+      {
+        switch_mutex_lock(a->mutex);
+        auto audioData = e.Result->GetAudioData();
+        for (size_t i = 0; i < audioData->size(); i += sizeof(int16_t)) {
+            int16_t value = static_cast<int16_t>((*audioData)[i]) | (static_cast<int16_t>((*audioData)[i + 1]) << 8);
+            pcm_data.push_back(value);
+        }
+
+        /* and write to the file */
+        size_t bytesResampled = pcm_data.size() * sizeof(uint16_t);
+        if (a->file) fwrite(pcm_data.data(), sizeof(uint16_t), pcm_data.size(), a->file);
+
+        // Resize the buffer if necessary
+        if (cBuffer->capacity() - cBuffer->size() < (bytesResampled / sizeof(uint16_t))) {
+
+          //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_SIZE));
+        }
+
+        /* Push the data into the buffer */
+        cBuffer->insert(cBuffer->end(), pcm_data.data(), pcm_data.data() + pcm_data.size());
+
+        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);
+          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, "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);
+            } 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");
+          }
+          switch_core_session_rwunlock(session);
+        }
+      }
+    };
+
+    speechSynthesizer->SynthesisCompleted += [a](const SpeechSynthesisEventArgs& e) {
+       switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts SynthesisCompleted\n");
+       a->draining = 1;
+    };
+
+    speechSynthesizer->SynthesisCanceled += [a](const SpeechSynthesisEventArgs& e) {
+      if (e.Result->Reason == ResultReason::Canceled) {
+        auto cancellation = SpeechSynthesisCancellationDetails::FromResult(e.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 synthsize tex %d with error string: %s.\n", static_cast<int>(cancellation->ErrorCode), cancellation->ErrorDetails.c_str());
+      }
+    };
+    // switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts before sending synthesize request\n");
+    if (std::strncmp(text, "<speak", 6) == 0) {
+      speechSynthesizer->SpeakSsmlAsync(text);
+    } else {
+      speechSynthesizer->SpeakTextAsync(text);
+    }
+    // switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "azure_speech_feed_tts sent synthesize request\n");
+    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;
+
+    {
+      switch_mutex_lock(a->mutex);
+      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;
+      }
+      if (cBuffer->empty()) {
+        if (a->draining) {
+          switch_mutex_unlock(a->mutex);
+          return SWITCH_STATUS_BREAK;
+        }
+        /* no audio available yet so send silence */
+        memset(data, 255, *datalen);
+        switch_mutex_unlock(a->mutex);
+        return SWITCH_STATUS_SUCCESS;
+      }
+      // azure returned 8000hz 16 bit data, we have to take enough data based on call sample rate.
+      size_t size = a->samples_rate ?
+        std::min((*datalen/(2 * a->samples_rate / 8000)), cBuffer->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(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 (a->cache_filename) {
+        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);
+        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");
+        }
+        switch_core_session_rwunlock(session);
+      }
+    }
+
+    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 +1,178 @@
+#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_INFO, "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_INFO, "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_INFO, "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_INFO, "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->samples_rate = 0;
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "a_speech_feed_tts\n");
+
+  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);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "a_speech_read_tts\n");
+  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);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_speech_flush_tts\n");
+  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_INFO, "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 +1,38 @@
+#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;
+  uint32_t samples_rate;
+
+  void *startTime;
+
+  FILE *file;
+  SpeexResamplerState *resampler;
+  void *circularBuffer;
+  switch_mutex_t *mutex;
+} 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_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/audio_pipe.cpp

@@ -48,13 +48,15 @@ int AudioPipe::lws_callback(struct lws *wsi,
         AudioPipe* ap = findPendingConnect(wsi);
         if (ap) {
           std::string apiKey = ap->getApiKey();
-          unsigned char **p = (unsigned char **)in, *end = (*p) + len;
-          char b[256];
-          memset(b, 0, sizeof(b));
-          strcpy(b,"Token ");
-          strcpy(b + 6, apiKey.c_str());
+          if (apiKey.length() > 0) {
+            unsigned char **p = (unsigned char **)in, *end = (*p) + len;
+            char b[256];
+            memset(b, 0, sizeof(b));
+            strcpy(b,"Token ");
+            strcpy(b + 6, apiKey.c_str());
 
-          if (lws_add_http_header_by_token(wsi, WSI_TOKEN_HTTP_AUTHORIZATION, (unsigned char *)b, strlen(b), p, end)) return -1;
+            if (lws_add_http_header_by_token(wsi, WSI_TOKEN_HTTP_AUTHORIZATION, (unsigned char *)b, strlen(b), p, end)) return -1;
+          }
         }
       }
       break;
@@ -255,6 +257,7 @@ static const lws_retry_bo_t retry = {
 };
 
 struct lws_context *AudioPipe::context = nullptr;
+std::thread AudioPipe::serviceThread;
 std::mutex AudioPipe::mutex_connects;
 std::mutex AudioPipe::mutex_disconnects;
 std::mutex AudioPipe::mutex_writes;
@@ -423,29 +426,35 @@ bool AudioPipe::lws_service_thread() {
 
 void AudioPipe::initialize(int loglevel, log_emit_function logger) {
 
-  lws_set_log_level(loglevel, logger);
+  //lws_set_log_level(loglevel, logger);
 
   lwsl_notice("AudioPipe::initialize starting\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
-  std::thread t(&deepgram::AudioPipe::lws_service_thread);
   stopFlag = false;
-  t.detach();
+  serviceThread = std::thread(&AudioPipe::lws_service_thread);
 }
 
 bool AudioPipe::deinitialize() {
   lwsl_notice("AudioPipe::deinitialize\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = true;
+  if (serviceThread.joinable()) {
+    serviceThread.join();
+  }
+
   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, const char* apiKey, notifyHandler_t callback) :
+  size_t bufLen, size_t minFreespace, const char* apiKey, int useTls, 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_audio_buffer_write_offset(LWS_PRE), m_recv_buf(nullptr), m_recv_buf_ptr(nullptr), 
-  m_state(LWS_CLIENT_IDLE), m_wsi(nullptr), m_vhd(nullptr), m_apiKey(apiKey), m_callback(callback) {
+  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) {
+
+  if (apiKey) m_apiKey = apiKey;
+  else m_apiKey = "";
 
   m_audio_buffer = new uint8_t[m_audio_buffer_max_len];
 }
@@ -470,7 +479,7 @@ bool AudioPipe::connect_client(struct lws_per_vhost_data *vhd) {
   i.path = m_path.c_str();
   i.host = i.address;
   i.origin = i.address;
-  i.ssl_connection = LCCSCF_USE_SSL;
+  if (m_useTls) i.ssl_connection = LCCSCF_USE_SSL;
   i.pwsi = &(m_wsi);
 
   m_state = LWS_CLIENT_CONNECTING;

mod_deepgram_transcribe/audio_pipe.hpp

@@ -45,7 +45,7 @@ namespace deepgram {
 
     // constructor
     AudioPipe(const char* uuid, const char* bugname, const char* host, unsigned int port, const char* path, 
-      size_t bufLen, size_t minFreespace, const char* apiKey, notifyHandler_t callback);
+      size_t bufLen, size_t minFreespace, const char* apiKey, int useTls, notifyHandler_t callback);
     ~AudioPipe();  
 
     LwsState_t getLwsState(void) { return m_state; }
@@ -86,6 +86,7 @@ namespace deepgram {
     void operator=(const AudioPipe&) = delete;
 
   private:
+    static std::thread serviceThread;
 
     static int lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
     static struct lws_context *context;
@@ -136,6 +137,7 @@ namespace deepgram {
     bool m_finished;
     std::string m_bugname;
     std::promise<void> m_promise;
+    bool m_useTls;
   };
 
 } // namespace deepgram

mod_deepgram_transcribe/dg_transcribe_glue.cpp

@@ -32,56 +32,59 @@ namespace {
   static unsigned int idxCallCount = 0;
   static uint32_t playCount = 0;
 
-  /* deepgram model / tier defaults by language */
-  struct LanguageInfo {
-      std::string tier;
-      std::string model;
+  /* deepgram model defaults by language */
+  static const std::unordered_map<std::string, std::string> languageLookupTable = {
+      {"zh", "base"},
+      {"zh-CN", "base"},
+      {"zh-TW", "base"},
+      {"da", "nova-2"},
+      {"da-DK", "nova-2"},
+      {"en", "nova-2"},
+      {"en-US", "nova-2"},
+      {"en-AU", "nova-2"},
+      {"en-GB", "nova-2"},
+      {"en-IN", "nova-2"},
+      {"en-NZ", "nova-2"},
+      {"nl", "nova-2"},
+      {"nl-BE", "nova-2"},
+      {"fr", "nova-2"},
+      {"fr-CA", "nova-2"},
+      {"de", "nova-2"},
+      {"el", "nova-2"},
+      {"hi", "nova-2"},
+      {"hi-Latn", "nova-2"},
+      {"id", "nova-2"},
+      {"it", "nova-2"},
+      {"ja", "enhanced"},
+      {"ko", "nova-2"},
+      {"ko-KR", "nova-2"},
+      {"no", "nova-2"},
+      {"pl", "nova-2"},
+      {"pt","nova-2"},
+      {"pt-BR", "nova-2"},
+      {"ru", "nova-2"},
+      {"es","nova-2"},
+      {"es-419","nova-2"},
+      {"es-LATAM","enhanced"},
+      {"sv", "nova-2"},
+      {"sv-SE", "nova-2"},
+      {"ta", "enhanced"},
+      {"taq", "enhanced"},
+      {"tr", "nova-2"},
+      {"uk", "nova-2"}
   };
 
-  static const std::unordered_map<std::string, LanguageInfo> languageLookupTable = {
-      {"zh", {"base", "general"}},
-      {"zh-CN", {"base", "general"}},
-      {"zh-TW", {"base", "general"}},
-      {"da", {"enhanced", "general"}},
-      {"en", {"nova", "phonecall"}},
-      {"en-US", {"nova", "phonecall"}},
-      {"en-AU", {"nova", "general"}},
-      {"en-GB", {"nova", "general"}},
-      {"en-IN", {"nova", "general"}},
-      {"en-NZ", {"nova", "general"}},
-      {"nl", {"enhanced", "general"}},
-      {"fr", {"enhanced", "general"}},
-      {"fr-CA", {"base", "general"}},
-      {"de", {"enhanced", "general"}},
-      {"hi", {"enhanced", "general"}},
-      {"hi-Latn", {"base", "general"}},
-      {"id", {"base", "general"}},
-      {"ja", {"enhanced", "general"}},
-      {"ko", {"enhanced", "general"}},
-      {"no", {"enhanced", "general"}},
-      {"pl", {"enhanced", "general"}},
-      {"pt", {"enhanced", "general"}},
-      {"pt-BR", {"enhanced", "general"}},
-      {"pt-PT", {"enhanced", "general"}},
-      {"ru", {"base", "general"}},
-      {"es", {"nova", "general"}},
-      {"es-419", {"nova", "general"}},
-      {"sv", {"enhanced", "general"}},
-      {"ta", {"enhanced", "general"}},
-      {"tr", {"base", "general"}},
-      {"uk", {"base", "general"}}
-  };
-
-  static bool getLanguageInfo(const std::string& language, LanguageInfo& info) {
+  static bool getLanguageInfo(const std::string& language, std::string& model) {
       auto it = languageLookupTable.find(language);
       if (it != languageLookupTable.end()) {
-          info = it->second;
+          model = it->second;
           return true;
       }
       return false;
   }
 
-  static const char* emptyTranscript = "{\"alternatives\":[{\"transcript\":\"\",\"confidence\":0.0,\"words\":[]}]}";
+  static const char* emptyTranscript = 
+    "\"is_final\":false,\"speech_final\":false,\"channel\":{\"alternatives\":[{\"transcript\":\"\",\"confidence\":0.0,\"words\":[]}]}";
 
   static void reaper(private_t *tech_pvt) {
     std::shared_ptr<deepgram::AudioPipe> pAp;
@@ -144,26 +147,22 @@ namespace {
     const char *var ;
     const char *model = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_MODEL");
     const char *customModel = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_CUSTOM_MODEL");
-    const char *tier = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_TIER") ;
     std::ostringstream oss;
-    LanguageInfo info;
 
     oss << "/v1/listen?";
 
-    if (!tier && !model && !customModel) {
-      /* make best choice by language */
-      if (getLanguageInfo(language, info)) {
-        oss << "tier=" << info.tier << "&model=" << info.model;
+    /* make best choice by language if model not supplied*/
+    if (!model && !customModel) {
+      std::string defaultModel;
+      if (getLanguageInfo(language, defaultModel)) {
+        oss << "&model=" << defaultModel;
       }
       else {
-        oss << "tier=base&model=general"; // most widely supported, though not ideal
+        oss << "model=base"; // most widely supported, though not ideal
       }
     }
-    else {
-      if (tier) oss << "tier=" << tier;
-      if (model) oss << "&model=" << model;
-      if (customModel) oss << "&model=" << customModel;
-    }
+    else if (model) oss << "&model=" << model;
+    else if (customModel) oss << "&model=" << customModel;
 
     if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_MODEL_VERSION")) {
      oss <<  "&version";
@@ -244,9 +243,6 @@ namespace {
      oss <<  "&tag=";
      oss <<  var;
     }
-    if (interim) {
-     oss <<  "&interim_results=true";
-    }
     if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_ENDPOINTING")) {
       oss <<  "&endpointing=";
       oss <<  var;
@@ -254,6 +250,11 @@ namespace {
     if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_UTTERANCE_END_MS")) {
       oss <<  "&utterance_end_ms=";
       oss <<  var;
+      interim = 1; 
+      //this requires interim_results=true (https://developers.deepgram.com/docs/understanding-end-of-speech-detection)
+    }
+    if (interim) {
+     oss <<  "&interim_results=true";
     }
     if (var = switch_channel_get_variable(channel, "DEEPGRAM_SPEECH_VAD_TURNOFF")) {
       oss <<  "&vad_turnoff=";
@@ -276,7 +277,7 @@ namespace {
         if (tech_pvt) {
           switch (event) {
             case deepgram::AudioPipe::CONNECT_SUCCESS:
-              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "connection successful\n");
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_INFO, "connection (%s) successful\n", tech_pvt->bugname);
               tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_CONNECT_SUCCESS, NULL, tech_pvt->bugname, finished);
             break;
             case deepgram::AudioPipe::CONNECT_FAIL:
@@ -286,19 +287,19 @@ namespace {
               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);
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_NOTICE, "connection (%s) failed: %s\n", message, tech_pvt->bugname);
             }
             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 dropped from far end\n");
+              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
               tech_pvt->pAudioPipe = nullptr;
-              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection closed gracefully\n");
+              switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "connection (%s) closed gracefully\n", tech_pvt->bugname);
             break;
             case deepgram::AudioPipe::MESSAGE:
               if( strstr(message, emptyTranscript)) {
@@ -306,7 +307,7 @@ namespace {
               }
               else {
                 tech_pvt->responseHandler(session, TRANSCRIBE_EVENT_RESULTS, message, tech_pvt->bugname, finished);
-                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "deepgram message: %s\n", message);
+                switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "deepgram message (%s): %s\n", tech_pvt->bugname, message);
               }
             break;
 
@@ -324,6 +325,7 @@ namespace {
     char* bugname, responseHandler_t responseHandler) {
 
     int err;
+    int useTls = true;
     switch_codec_implementation_t read_impl;
     switch_channel_t *channel = switch_core_session_get_channel(session);
 
@@ -336,8 +338,32 @@ namespace {
     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);
-    strncpy(tech_pvt->host, "api.deepgram.com", MAX_WS_URL_LEN);
-    tech_pvt->port = 443;
+
+    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(':');
+      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;
+      }
+      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;
+    }
+
     strncpy(tech_pvt->path, path.c_str(), MAX_PATH_LEN);    
     tech_pvt->sampling = desiredSampling;
     tech_pvt->responseHandler = responseHandler;
@@ -350,13 +376,13 @@ namespace {
 
     const char* apiKey = switch_channel_get_variable(channel, "DEEPGRAM_API_KEY");
     if (!apiKey && defaultApiKey) apiKey = defaultApiKey;
-    else if (!apiKey) {
+    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, eventCallback);
+      buflen, read_impl.decoded_bytes_per_packet, apiKey, useTls, eventCallback);
     if (!ap) {
       switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_ERROR, "Error allocating AudioPipe\n");
       return SWITCH_STATUS_FALSE;

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 -shared `pkg-config --libs boost` -lstdc++ -lmpg123

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 +1,663 @@
+#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);
+
+  curl_easy_setopt(_easy, CURLOPT_HTTP_VERSION, 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 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, int gain, bool loop) {
+  _method = HttpMethod_t::HTTP_METHOD_POST;
+  _url = url;
+  _body = body;
+  _headers = headers;
+  _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 +1,134 @@
+#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, 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::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;
+    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) != 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, 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 +1,331 @@
+/* 
+ *
+ * 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;
+  
+  if (bug) {
+    struct cap_cb *cb =(struct cap_cb *) switch_core_media_bug_get_user_data(bug);
+    status = silence_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);
+      return SWITCH_STATUS_FALSE;
+    }
+  }
+  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 > 4 ? atoi(argv[4]) : 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 +1,143 @@
+#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) {
+    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);
+    _apQueue.pop();
+    if (!_apQueue.empty()) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "onPlayDone: starting queued audio on track %s\n", _trackName.c_str());
+      _apQueue.front()->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;
+  }
+
+  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::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, int gain, bool loop) {
+  bool startIt = false;
+  if (_stopping) return;
+  auto ap = std::make_shared<AudioProducerHttp>(_mutex, _buffer, _sampleRate);
+    ap->queueHttpPostAudio(url, body, headers, 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, 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 +1,118 @@
+#include "tts_vendor_parser.h"
+
+#include <sstream>
+#include <switch_json.h>
+#include <map>
+
+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) {
+  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 {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "tts_vendor_parse_text: There is no available parser for text\n");
+    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);
+
+#endif

mod_dub/vector_math.cpp

@@ -0,0 +1,211 @@
+#include "vector_math.h"
+#include <assert.h>
+#include <string.h>
+
+#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
+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/Makefile.am

@@ -0,0 +1,8 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_elevenlabs_tts
+
+mod_LTLIBRARIES = mod_elevenlabs_tts.la
+mod_elevenlabs_tts_la_SOURCES  = mod_elevenlabs_tts.c elevenlabs_glue.cpp
+mod_elevenlabs_tts_la_CFLAGS   = $(AM_CFLAGS)
+mod_elevenlabs_tts_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_elevenlabs_tts_la_LDFLAGS  = -avoid-version -module -no-undefined -shared `pkg-config --libs boost` -lstdc++

mod_elevenlabs_tts/README.md

@@ -0,0 +1,23 @@
+# mod_google_tts
+
+A Freeswitch module that allows Google 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)
+
+### Events
+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'
+  });
+```
+## Examples
+[google_tts.js](../../examples/google_tts.js)

mod_elevenlabs_tts/elevenlabs_glue.h

@@ -0,0 +1,12 @@
+#ifndef __ELEVENLABS_GLUE_H__
+#define __ELEVENLABS_GLUE_H__
+
+switch_status_t elevenlabs_speech_load();
+switch_status_t elevenlabs_speech_open(elevenlabs_t* elevenlabs);
+switch_status_t elevenlabs_speech_feed_tts(elevenlabs_t* elevenlabs, char* text, switch_speech_flag_t *flags);
+switch_status_t elevenlabs_speech_read_tts(elevenlabs_t* elevenlabs, void *data, size_t *datalen, switch_speech_flag_t *flags);
+switch_status_t elevenlabs_speech_flush_tts(elevenlabs_t* elevenlabs);
+switch_status_t elevenlabs_speech_close(elevenlabs_t* elevenlabs);
+switch_status_t elevenlabs_speech_unload();
+
+#endif

mod_elevenlabs_tts/mod_elevenlabs_tts.c

@@ -0,0 +1,214 @@
+/* 
+ *
+ * mod_elevenlabs_tts.c -- Google GRPC-based text to speech
+ *
+ */
+#include "mod_elevenlabs_tts.h"
+#include "elevenlabs_glue.h"
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_elevenlabs_tts_load);
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_elevenlabs_tts_shutdown);
+SWITCH_MODULE_DEFINITION(mod_elevenlabs_tts, mod_elevenlabs_tts_load, mod_elevenlabs_tts_shutdown, NULL);
+
+static void clearElevenlabs(elevenlabs_t* el, int freeAll) {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "clearElevenlabs\n");
+  if (el->api_key) free(el->api_key);
+  if (el->model_id) free(el->model_id);
+  if (el->similarity_boost) free(el->similarity_boost);
+  if (el->stability) free(el->stability);
+  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->ct) free(el->ct);
+  if (el->reported_latency) free(el->reported_latency);
+  if (el->request_id) free(el->request_id);
+  if (el->history_item_id) free(el->history_item_id);
+  if (el->err_msg) free(el->err_msg);
+  if (el->name_lookup_time_ms) free(el->name_lookup_time_ms);
+  if (el->connect_time_ms) free(el->connect_time_ms);
+  if (el->final_response_time_ms) free(el->final_response_time_ms);
+  if (el->cache_filename) free(el->cache_filename);
+  
+  el->api_key = NULL;
+  el->model_id = NULL;
+  el->similarity_boost = NULL;
+  el->stability = NULL;
+  el->style = NULL;
+  el->use_speaker_boost = NULL;
+  el->optimize_streaming_latency = NULL;
+  el->ct = NULL;
+  el->reported_latency = NULL;
+  el->request_id = NULL;
+  el->history_item_id = NULL;
+  el->err_msg = NULL;
+  el->name_lookup_time_ms = NULL;
+  el->connect_time_ms = NULL;
+  el->final_response_time_ms = NULL;
+  el->cache_filename = NULL;
+
+  el->file = NULL;
+
+  if (freeAll) {
+    if (el->voice_name) free(el->voice_name);
+    if (el->session_id) free(el->session_id);
+    el->voice_name = NULL;
+    el->session_id = NULL;
+  }
+}
+static elevenlabs_t * createOrRetrievePrivateData(switch_speech_handle_t *sh) {
+  elevenlabs_t *el = (elevenlabs_t *) sh->private_info;  
+  if (!el) {
+    el = switch_core_alloc(sh->memory_pool, sizeof(*el));
+  	sh->private_info = el;
+    memset(el, 0, sizeof(*el));
+    switch_mutex_init(&el->mutex, SWITCH_MUTEX_NESTED, sh->memory_pool);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "allocated elevenlabs_t\n");
+  }
+  return el;
+}
+
+/**
+ * this is called when FreeSWITCH loads the module
+ * if the module is retrieved from cache this will not be called
+ * therefore, probably best not to connect to elevenlabs yet
+ * Note: if cached, the voice will be provide as a param
+ */
+static switch_status_t ell_speech_open(switch_speech_handle_t *sh, const char *voice_name, int rate, int channels, switch_speech_flag_t *flags)
+{
+
+  elevenlabs_t *el = createOrRetrievePrivateData(sh);
+  el->voice_name = strdup(voice_name);
+  el->rate = rate;
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "ell_speech_open voice: %s, rate %d, channels %d\n", voice_name, rate, channels);
+	return elevenlabs_speech_open(el);
+}
+
+static switch_status_t ell_speech_close(switch_speech_handle_t *sh, switch_speech_flag_t *flags)
+{
+  switch_status_t rc;
+	elevenlabs_t *el = (elevenlabs_t *) sh->private_info;
+	assert(el != NULL);
+  
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "ell_speech_close\n");
+
+  switch_mutex_destroy(el->mutex);
+
+	rc = elevenlabs_speech_close(el);
+  clearElevenlabs(el, 1);
+  return rc;  
+}
+
+/**
+ * Freeswitch will call this function to feed us text to speak
+ */
+static switch_status_t ell_speech_feed_tts(switch_speech_handle_t *sh, char *text, switch_speech_flag_t *flags)
+{
+  elevenlabs_t *el = createOrRetrievePrivateData(sh);
+  el->draining = 0;
+  el->reads = 0;
+  
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "ell_speech_feed_tts\n");
+  
+	return elevenlabs_speech_feed_tts(el, text, flags);
+}
+
+/**
+ * This is called at the end, not sure exactly what we need to do here..
+ */
+static void ell_speech_flush_tts(switch_speech_handle_t *sh)
+{
+	elevenlabs_t *el = (elevenlabs_t *) sh->private_info;
+	//assert(el != NULL);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "ell_speech_flush_tts\n");
+  elevenlabs_speech_flush_tts(el);
+
+  clearElevenlabs(el, 0);
+}
+
+/**
+ * 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 ell_speech_read_tts(switch_speech_handle_t *sh, void *data, size_t *datalen, switch_speech_flag_t *flags)
+{
+	elevenlabs_t *el = (elevenlabs_t *) sh->private_info;
+  return elevenlabs_speech_read_tts(el, data, datalen, flags);
+}
+
+static void ell_text_param_tts(switch_speech_handle_t *sh, char *param, const char *val)
+{
+  elevenlabs_t *el = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "ell_text_param_tts: %s=%s\n", param, val);
+
+  if (0 == strcmp(param, "voice")) {
+    if (el->voice_name) free(el->voice_name);
+    el->voice_name = strdup(val);
+  }
+  else if (0 == strcmp(param, "api_key")) {
+    if (el->api_key) free(el->api_key);
+    el->api_key = strdup(val);
+  }
+  else if (0 == strcmp(param, "model_id")) {
+    if (el->model_id) free(el->model_id);
+    el->model_id = strdup(val);
+  }
+  else if (0 == strcmp(param, "similarity_boost")) {
+    if (el->similarity_boost) free(el->similarity_boost);
+    el->similarity_boost = strdup(val);
+  }
+  else if (0 == strcmp(param, "stability")) {
+    if (el->stability) free(el->stability);
+    el->stability = strdup(val);
+  }
+  else if (0 == strcmp(param, "style")) {
+    if (el->style) free(el->style);
+    el->style = strdup(val);
+  }
+  else if (0 == strcmp(param, "use_speaker_boost")) {
+    if (el->use_speaker_boost) free(el->use_speaker_boost);
+    el->use_speaker_boost = strdup(val);
+  }
+  else if (0 == strcmp(param, "optimize_streaming_latency")) {
+    if (el->optimize_streaming_latency) free(el->optimize_streaming_latency);
+    el->optimize_streaming_latency = strdup(val);
+  }
+  else if (0 == strcmp(param, "write_cache_file") && switch_true(val)) {
+    el->cache_audio = 1;
+  }
+  else if (0 == strcmp(param, "session-uuid")) {
+    if (el->session_id) free(el->session_id);
+    el->session_id = strdup(val);
+  }
+}
+
+static void ell_numeric_param_tts(switch_speech_handle_t *sh, char *param, int val)
+{
+}
+
+static void ell_float_param_tts(switch_speech_handle_t *sh, char *param, double val)
+{
+}
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_elevenlabs_tts_load)
+{
+	switch_speech_interface_t *speech_interface;
+
+	/* connect my internal structure to the blank pointer passed to me */
+	*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 = "elevenlabs";
+	speech_interface->speech_open = ell_speech_open;
+	speech_interface->speech_close = ell_speech_close;
+	speech_interface->speech_feed_tts = ell_speech_feed_tts;
+	speech_interface->speech_read_tts = ell_speech_read_tts;
+	speech_interface->speech_flush_tts = ell_speech_flush_tts;
+	speech_interface->speech_text_param_tts = ell_text_param_tts;
+	speech_interface->speech_numeric_param_tts = ell_numeric_param_tts;
+	speech_interface->speech_float_param_tts = ell_float_param_tts;
+
+	return elevenlabs_speech_load();
+}
+
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_elevenlabs_tts_shutdown)
+{
+  return elevenlabs_speech_unload();
+}

mod_elevenlabs_tts/mod_elevenlabs_tts.h

@@ -0,0 +1,48 @@
+#ifndef __MOD_ELEVENLABS_TTS_H__
+#define __MOD_ELEVENLABS_TTS_H__
+
+#include <switch.h>
+#include <fcntl.h>
+#include <sys/stat.h>
+#include <unistd.h>
+#include <speex/speex_resampler.h>
+
+struct elevenlabs_data {
+  char *session_id;
+	char *voice_name;
+  char* api_key;
+  char* model_id;
+  char* similarity_boost;
+  char* stability;
+  char* style;
+  char* use_speaker_boost;
+  char* optimize_streaming_latency;
+
+  /* result data */
+  long response_code;
+  char *ct;
+  char *reported_latency;
+  char *request_id;
+  char *history_item_id;
+  char *name_lookup_time_ms;
+  char *connect_time_ms;
+  char *final_response_time_ms;
+  char *err_msg;
+  char *cache_filename;
+
+	int rate;
+  uint32_t sample_rate;
+
+  void *conn;
+	FILE *file;
+  switch_mutex_t *mutex;
+  void *circularBuffer;
+  int draining;
+  int reads;
+  int cache_audio;
+  SpeexResamplerState *resampler;
+};
+
+typedef struct elevenlabs_data elevenlabs_t;
+
+#endif

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 +1,381 @@
+#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();
+      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;
+}
+
+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 +1,339 @@
+#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::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);
+            }
+        }
+    }
+
+    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 orr 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();
+      }
+    }
+    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);
+        }
+      }
+      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) {
@@ -69,19 +111,19 @@ 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 +133,29 @@ 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 +166,27 @@ 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 +194,7 @@ 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 +203,8 @@ 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 +216,7 @@ 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 +226,15 @@ 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 +250,23 @@ 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 +283,11 @@ 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 +325,16 @@ 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 +344,7 @@ 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 +356,9 @@ 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 +375,24 @@ 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 +408,9 @@ 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 +435,9 @@ 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 +454,28 @@ 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);
 		}
@@ -481,4 +564,3 @@ 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

@@ -37,6 +37,10 @@ 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_ibm_transcribe/audio_pipe.cpp

@@ -48,7 +48,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
         lwsl_err("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(), AudioPipe::CONNECT_FAIL, (char *) in, ap->isFinished(), ap->isInterimTranscriptsEnabled(), ap->getBugname().c_str());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_FAIL, (char *) in, ap->isFinished(), ap->isInterimTranscriptsEnabled());
         }
         else {
           lwsl_err("AudioPipe::lws_service_thread LWS_CALLBACK_CLIENT_CONNECTION_ERROR unable to find wsi %p..\n", wsi); 
@@ -77,7 +77,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
             oss << "}";
 
             ap->bufferForSending(oss.str().c_str());
-            ap->m_callback(ap->m_uuid.c_str(), AudioPipe::CONNECT_SUCCESS, NULL,  ap->isFinished(), ap->isInterimTranscriptsEnabled(), ap->getBugname().c_str());
+            ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECT_SUCCESS, NULL,  ap->isFinished(), ap->isInterimTranscriptsEnabled());
           }
 
         }
@@ -97,12 +97,12 @@ int AudioPipe::lws_callback(struct lws *wsi,
           // closed by us
 
           lwsl_debug("%s socket closed by us\n", ap->m_uuid.c_str());
-          ap->m_callback(ap->m_uuid.c_str(), AudioPipe::CONNECTION_CLOSED_GRACEFULLY, NULL,  ap->isFinished(), ap->isInterimTranscriptsEnabled(), ap->getBugname().c_str());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_CLOSED_GRACEFULLY, NULL,  ap->isFinished(), ap->isInterimTranscriptsEnabled());
         }
         else if (ap->m_state == LWS_CLIENT_CONNECTED) {
           // closed by far end
           lwsl_info("%s socket closed by far end\n", ap->m_uuid.c_str());
-          ap->m_callback(ap->m_uuid.c_str(), AudioPipe::CONNECTION_DROPPED, NULL,  ap->isFinished(), ap->isInterimTranscriptsEnabled(), ap->getBugname().c_str());
+          ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::CONNECTION_DROPPED, NULL,  ap->isFinished(), ap->isInterimTranscriptsEnabled());
         }
         ap->m_state = LWS_CLIENT_DISCONNECTED;
         ap->setClosed();
@@ -166,7 +166,7 @@ int AudioPipe::lws_callback(struct lws *wsi,
               std::string msg((char *)ap->m_recv_buf, ap->m_recv_buf_ptr - ap->m_recv_buf);
               //std::cerr << "Recv: " << msg << std::endl;
 
-              ap->m_callback(ap->m_uuid.c_str(), AudioPipe::MESSAGE, msg.c_str(),  ap->isFinished(), ap->isInterimTranscriptsEnabled(), ap->getBugname().c_str());
+              ap->m_callback(ap->m_uuid.c_str(), ap->m_bugname.c_str(), AudioPipe::MESSAGE, msg.c_str(),  ap->isFinished(), ap->isInterimTranscriptsEnabled());
               if (nullptr != ap->m_recv_buf) free(ap->m_recv_buf);
             }
             ap->m_recv_buf = ap->m_recv_buf_ptr = nullptr;
@@ -251,6 +251,7 @@ static const lws_retry_bo_t retry = {
 };
 
 struct lws_context *AudioPipe::context = nullptr;
+std::thread AudioPipe::serviceThread;
 std::mutex AudioPipe::mutex_connects;
 std::mutex AudioPipe::mutex_disconnects;
 std::mutex AudioPipe::mutex_writes;
@@ -418,26 +419,29 @@ bool AudioPipe::lws_service_thread() {
 
 void AudioPipe::initialize(int loglevel, log_emit_function logger) {
 
-  lws_set_log_level(loglevel, logger);
+  //lws_set_log_level(loglevel, logger);
 
   lwsl_notice("AudioPipe::initialize starting\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
-  std::thread t(&AudioPipe::lws_service_thread);
   stopFlag = false;
-  t.detach();
+  serviceThread = std::thread(&AudioPipe::lws_service_thread);
 }
 
 bool AudioPipe::deinitialize() {
   lwsl_notice("AudioPipe::deinitialize\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = true;
+  if (serviceThread.joinable()) {
+    serviceThread.join();
+  }
+
   return true;
 }
 
 // instance members
-AudioPipe::AudioPipe(const char* uuid, const char* host, unsigned int port, const char* path,
+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_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_interim(false),
   m_state(LWS_CLIENT_IDLE), m_wsi(nullptr), m_vhd(nullptr), m_callback(callback) {

mod_ibm_transcribe/audio_pipe.hpp

@@ -31,7 +31,7 @@ public:
     MESSAGE
   };
   typedef void (*log_emit_function)(int level, const char *line);
-  typedef void (*notifyHandler_t)(const char *sessionId, NotifyEvent_t event, const char* message, bool finished, bool wantsInterim, const char* bugname);
+  typedef void (*notifyHandler_t)(const char *sessionId, const char* bugname, NotifyEvent_t event, const char* message, bool finished, bool wantsInterim);
 
   struct lws_per_vhost_data {
     struct lws_context *context;
@@ -44,7 +44,7 @@ public:
   static bool lws_service_thread();
 
   // constructor
-  AudioPipe(const char* uuid, const char* host, unsigned int port, const char* path, 
+  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();  
 
@@ -104,6 +104,7 @@ public:
   void operator=(const AudioPipe&) = delete;
 
 private:
+    static std::thread serviceThread;
 
   static int lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
   static struct lws_context *context;

mod_ibm_transcribe/ibm_transcribe_glue.cpp

@@ -82,8 +82,8 @@ namespace {
     }
   }
 
-  static void responseHandler(switch_core_session_t* session, 
-    const char* eventName, const char * json, const char* bugname, int finished) {
+  static void responseHandler(switch_core_session_t* session, const char* bugname,
+    const char* eventName, const char * json, int finished) {
     switch_event_t *event;
     switch_channel_t *channel = switch_core_session_get_channel(session);
 
@@ -162,7 +162,7 @@ namespace {
     return path;
   }
 
-  static void eventCallback(const char* sessionId, ibm::AudioPipe::NotifyEvent_t event, const char* message, bool finished, bool wantsInterim, const char* bugname) {
+  static void eventCallback(const char* sessionId, const char* bugname, ibm::AudioPipe::NotifyEvent_t event, const char* message, bool finished, bool wantsInterim) {
     switch_core_session_t* session = switch_core_session_locate(sessionId);
     if (session) {
       bool releaseAudioPipe = false;
@@ -256,10 +256,11 @@ namespace {
     tech_pvt->channels = channels;
     tech_pvt->id = ++idxCallCount;
     tech_pvt->buffer_overrun_notified = 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);
 
-    ibm::AudioPipe* ap = new ibm::AudioPipe(tech_pvt->sessionId, tech_pvt->host, tech_pvt->port, tech_pvt->path, 
+    ibm::AudioPipe* ap = new ibm::AudioPipe(tech_pvt->sessionId, 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");
@@ -311,7 +312,8 @@ extern "C" {
   switch_status_t ibm_transcribe_init() {
     switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_ibm_transcribe: audio buffer (in secs):    %d secs\n", nAudioBufferSecs);
  
-    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 ;
     
     ibm::AudioPipe::initialize(logs, lws_logger);
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AudioPipe::initialize completed\n");

mod_ibm_transcribe/mod_ibm_transcribe.c

@@ -56,7 +56,7 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	void *pUserData;
 	uint32_t samples_per_second;
 
-	if (switch_channel_get_private(channel, MY_BUG_NAME)) {
+	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);
 	}
@@ -76,7 +76,7 @@ static switch_status_t start_capture(switch_core_session_t *session, switch_medi
 	if ((status = switch_core_media_bug_add(session, "ibm_transcribe", NULL, capture_callback, pUserData, 0, flags, &bug)) != SWITCH_STATUS_SUCCESS) {
 		return status;
 	}
-  switch_channel_set_private(channel, MY_BUG_NAME, bug);
+  switch_channel_set_private(channel, bugname, bug);
 	switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "added media bug for ibm transcribe\n");
 
 	return SWITCH_STATUS_SUCCESS;
@@ -87,7 +87,7 @@ 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, MY_BUG_NAME);
+	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.\n");

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

@@ -240,6 +240,7 @@ static const lws_retry_bo_t retry = {
 };
 
 struct lws_context *AudioPipe::context = nullptr;
+std::thread AudioPipe::serviceThread;
 std::mutex AudioPipe::mutex_connects;
 std::mutex AudioPipe::mutex_disconnects;
 std::mutex AudioPipe::mutex_writes;
@@ -408,19 +409,21 @@ bool AudioPipe::lws_service_thread() {
 
 void AudioPipe::initialize(int loglevel, log_emit_function logger) {
 
-  lws_set_log_level(loglevel, logger);
+  //lws_set_log_level(loglevel, logger);
 
   lwsl_notice("AudioPipe::initialize starting\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
-  std::thread t(&AudioPipe::lws_service_thread);
   stopFlag = false;
-  t.detach();
+  serviceThread = std::thread(&AudioPipe::lws_service_thread);
 }
 
 bool AudioPipe::deinitialize() {
   lwsl_notice("AudioPipe::deinitialize\n"); 
   std::lock_guard<std::mutex> lock(mapMutex);
   stopFlag = true;
+  if (serviceThread.joinable()) {
+    serviceThread.join();
+  }
   return true;
 }
 

mod_jambonz_transcribe/audio_pipe.hpp

@@ -86,6 +86,7 @@ public:
   void operator=(const AudioPipe&) = delete;
 
 private:
+    static std::thread serviceThread;
 
   static int lws_callback(struct lws *wsi, enum lws_callback_reasons reason, void *user, void *in, size_t len); 
   static struct lws_context *context;

mod_jambonz_transcribe/jb_transcribe_glue.cpp

@@ -256,7 +256,6 @@ namespace {
     switch_log_printf(SWITCH_CHANNEL_SESSION_LOG(session), SWITCH_LOG_DEBUG, "host: %s, port: %d, path: %s\n", host, port, path);
 
     strncpy(tech_pvt->sessionId, switch_core_session_get_uuid(session), MAX_SESSION_ID);
-    strncpy(tech_pvt->bugname, bugname, MAX_BUG_LEN);
     strncpy(tech_pvt->host, host, MAX_WS_URL_LEN);
     strncpy(tech_pvt->path, path, MAX_PATH_LEN); 
     tech_pvt->port = port;
@@ -268,6 +267,7 @@ namespace {
     tech_pvt->channels = channels;
     tech_pvt->id = ++idxCallCount;
     tech_pvt->buffer_overrun_notified = 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);
 
@@ -325,7 +325,8 @@ extern "C" {
   switch_status_t jb_transcribe_init() {
     switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_jambonz_transcribe: audio buffer (in secs):    %d secs\n", nAudioBufferSecs);
  
-    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 ;
     
     jambonz::AudioPipe::initialize(logs, lws_logger);
 		switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "AudioPipe::initialize completed\n");

mod_nuance_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_nvidia_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_soniox_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_whisper_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_whisper_tts/Makefile.am

@@ -0,0 +1,8 @@
+include $(top_srcdir)/build/modmake.rulesam
+MODNAME=mod_whisper_tts
+
+mod_LTLIBRARIES = mod_whisper_tts.la
+mod_whisper_tts_la_SOURCES  = mod_whisper_tts.c whisper_glue.cpp
+mod_whisper_tts_la_CFLAGS   = $(AM_CFLAGS)
+mod_whisper_tts_la_LIBADD   = $(switch_builddir)/libfreeswitch.la
+mod_whisper_tts_la_LDFLAGS  = -avoid-version -module -no-undefined -shared `pkg-config --libs boost` -lstdc++ -lmpg123

mod_whisper_tts/README.md

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

mod_whisper_tts/mod_whisper_tts.c

@@ -0,0 +1,164 @@
+#include "mod_whisper_tts.h"
+#include "whisper_glue.h"
+
+SWITCH_MODULE_LOAD_FUNCTION(mod_whisper_tts_load);
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_whisper_tts_shutdown);
+SWITCH_MODULE_DEFINITION(mod_whisper_tts, mod_whisper_tts_load, mod_whisper_tts_shutdown, NULL);
+
+static void clearWhisper(whisper_t* w, int freeAll) {
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "clearWhisper\n");
+  if (w->api_key) free(w->api_key);
+  if (w->model_id) free(w->model_id);
+  if (w->speed) free(w->speed);
+  if (w->request_id) free(w->request_id);
+  if (w->reported_latency) free(w->reported_latency);
+  if (w->ct) free(w->ct);
+  if (w->err_msg) free(w->err_msg);
+  if (w->name_lookup_time_ms) free(w->name_lookup_time_ms);
+  if (w->connect_time_ms) free(w->connect_time_ms);
+  if (w->final_response_time_ms) free(w->final_response_time_ms);
+  if (w->cache_filename) free(w->cache_filename);
+  
+
+  w->api_key = NULL;
+  w->model_id = NULL;
+  w->speed = NULL;
+  w->request_id = NULL;
+  w->reported_latency = NULL;
+  w->ct = NULL;
+  w->err_msg = NULL;
+  w->name_lookup_time_ms = NULL;
+  w->connect_time_ms = NULL;
+  w->final_response_time_ms = NULL;
+  w->cache_filename = NULL;
+
+  if (freeAll) {
+    if (w->voice_name) free(w->voice_name);
+    if (w->session_id) free(w->session_id);
+    w->voice_name = NULL;
+    w->session_id = NULL;
+  }
+}
+
+static whisper_t * createOrRetrievePrivateData(switch_speech_handle_t *sh) {
+  whisper_t *w = (whisper_t *) sh->private_info;  
+  if (!w) {
+    w = switch_core_alloc(sh->memory_pool, sizeof(*w));
+  	sh->private_info = w;
+    memset(w, 0, sizeof(*w));
+    switch_mutex_init(&w->mutex, SWITCH_MUTEX_NESTED, sh->memory_pool);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "allocated whisper_t\n");
+  }
+  return w;
+}
+
+switch_status_t w_speech_open(switch_speech_handle_t *sh, const char *voice_name, int rate, int channels, switch_speech_flag_t *flags)
+{
+  whisper_t *w = createOrRetrievePrivateData(sh);
+  w->voice_name = strdup(voice_name);
+  w->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 whisper_speech_open(w);
+}
+
+static switch_status_t w_speech_close(switch_speech_handle_t *sh, switch_speech_flag_t *flags)
+{
+  switch_status_t rc;
+  whisper_t *w = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_speech_close\n");
+
+  switch_mutex_destroy(w->mutex);
+
+  rc = whisper_speech_close(w);
+  clearWhisper(w, 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)
+{
+  whisper_t *w = createOrRetrievePrivateData(sh);
+  w->draining = 0;
+  w->reads = 0;
+
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_speech_feed_tts\n");
+
+  return whisper_speech_feed_tts(w, 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)
+{
+  whisper_t *w = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_speech_read_tts\n");
+  return whisper_speech_read_tts(w, 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)
+{
+  whisper_t *w = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_speech_flush_tts\n");
+  whisper_speech_flush_tts(w);
+
+  clearWhisper(w, 0);
+}
+
+static void w_text_param_tts(switch_speech_handle_t *sh, char *param, const char *val)
+{
+  whisper_t *w = createOrRetrievePrivateData(sh);
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "w_text_param_tts: %s=%s\n", param, val);
+  if (0 == strcmp(param, "api_key")) {
+    if (w->api_key) free(w->api_key);
+    w->api_key = strdup(val);
+  } else if (0 == strcmp(param, "voice")) {
+    if (w->voice_name) free(w->voice_name);
+    w->voice_name = strdup(val);
+  } else if (0 == strcmp(param, "model_id")) {
+    if (w->model_id) free(w->model_id);
+    w->model_id = strdup(val);
+  } else if (0 == strcmp(param, "speed")) {
+    if (w->speed) free(w->speed);
+    w->speed = strdup(val);
+  } else if (0 == strcmp(param, "session-uuid")) {
+    if (w->session_id) free(w->session_id);
+    w->session_id = strdup(val);
+  } else if (0 == strcmp(param, "write_cache_file") && switch_true(val)) {
+    w->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_whisper_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 = "whisper";
+  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 whisper_speech_load();
+}
+
+SWITCH_MODULE_SHUTDOWN_FUNCTION(mod_whisper_tts_shutdown)
+{
+  return whisper_speech_unload();
+}

mod_whisper_tts/mod_whisper_tts.h

@@ -0,0 +1,33 @@
+#ifndef __MOD_WHISPER_TTS_H__
+#define __MOD_WHISPER_TTS_H__
+
+#include <switch.h>
+typedef struct whisper_data {
+  char *voice_name;
+  char *api_key;
+  char *model_id;
+  char *speed;
+
+  /* result data */
+  long response_code;
+  char *ct;
+  char *reported_latency;
+  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;
+
+	void *conn;
+  void *circularBuffer;
+  switch_mutex_t *mutex;
+  FILE *file;
+} whisper_t;
+#endif

mod_whisper_tts/whisper_glue.cpp

@@ -0,0 +1,964 @@
+#include "mod_whisper_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;
+  whisper_t* whisper;
+  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 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, 10L);
+
+  return easy ;    
+}
+
+static void cleanupConn(ConnInfo_t *conn) {
+  auto w = conn->whisper;
+
+  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 ;
+  }
+
+  w->conn = nullptr ;
+  w->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 w = conn->whisper;
+      w->response_code = response_code;
+      if (ct) w->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_whisper_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());
+
+      w->name_lookup_time_ms = strdup(name_lookup_ms.c_str());
+      w->connect_time_ms = strdup(connect_ms.c_str());
+      w->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_whisper_tts threadFunc - starting\n");
+
+  for(;;) {
+      
+    try {
+      io_service.run() ;
+      break ;
+    }
+    catch( std::exception& e) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "mod_whisper_tts threadFunc - Error: %s\n", e.what());
+    }
+  }
+  switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "mod_whisper_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 w = conn->whisper;
+  CircularBuffer_t *cBuffer = (CircularBuffer_t *) w->circularBuffer;
+  std::vector<uint16_t> pcm_data;
+  
+  if (conn->flushed) {
+    /* this will abort the transfer */
+    return 0;
+  }
+  {
+    switch_mutex_lock(w->mutex);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "write_cb: received data, response %ld\n", 
+      w->response_code);
+
+    if (w->response_code > 0 && w->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());
+      w->err_msg = strdup(body.c_str());
+      switch_mutex_unlock(w->mutex);
+      return 0;
+    }
+
+    /* cache file will stay in the mp3 format for size (smaller) and simplicity */
+    if (conn->file) fwrite(data, sizeof(uint8_t), bytes_received, conn->file);
+
+    pcm_data = convert_mp3_to_linear(conn, data, bytes_received);
+    size_t bytesResampled = pcm_data.size() * sizeof(uint16_t);
+
+    // 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 == w->reads++) {
+      fireEvent = true;
+    }
+    switch_mutex_unlock(w->mutex);
+  }
+  if (fireEvent && w->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(w->session_id);
+    if (session) {
+      switch_channel_t *channel = switch_core_session_get_channel(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 (w->reported_latency) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_reported_latency_ms", w->reported_latency);
+          }
+          if (w->request_id) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_request_id", w->request_id);
+          }
+          if (w->name_lookup_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_name_lookup_time_ms", w->name_lookup_time_ms);
+          }
+          if (w->connect_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_connect_time_ms", w->connect_time_ms);
+          }
+          if (w->final_response_time_ms) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_final_response_time_ms", w->final_response_time_ms);
+          }
+          if (w->voice_name) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_voice_name", w->voice_name);
+          }
+          if (w->model_id) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_whisper_model_id", w->model_id);
+          }
+          if (w->cache_filename) {
+            switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", w->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");
+      }
+      switch_core_session_rwunlock(session);
+    }
+    else {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "write_cb: session %s not found\n", w->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 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 ";
+  whisper_t* w = conn->whisper;
+  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("openai-processing-ms")) w->reported_latency = strdup(value.c_str());
+    else if (0 == header.compare("x-request-id")) w->request_id = strdup(value.c_str());
+  }
+  else {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "recv header: %s\n", input.c_str());
+    if (input.rfind(prefix, 0) == 0) {
+      try {
+        w->response_code = std::stoi(input.substr(prefix.length()));
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "parsed response code: %ld\n", w->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 whisper_speech_load() {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "whisper_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, "whisper_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 = "/var/";
+    }
+    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";
+
+    // 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, "whisper_speech_loaded..\n");
+
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t whisper_speech_unload() {
+    /* stop the ASIO IO service */
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "whisper_speech_unload: stopping io service\n");
+    io_service.stop();
+
+    /* Join the worker thread */
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "whisper_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, "whisper_speech_unload: release curl multi\n");
+    curl_multi_cleanup(global.multi);
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_NOTICE, "whisper_speech_unload: completed\n");
+
+    mpg123_exit();
+
+		return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t whisper_speech_open(whisper_t* whisper) {
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t whisper_speech_feed_tts(whisper_t* w, 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 (w->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);
+      w->cache_filename = strdup(outfile);
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "writing audio cache file to %s\n", w->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 {
+        w->file = fdopen(fd, "wb");
+        if (!w->file) {
+          close(fd);
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error opening cache file %s: %s\n", outfile, strerror(errno));
+        }
+      }
+    }
+
+    if (!w->api_key) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "whisper_speech_feed_tts: no api_key provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+    if (!w->model_id) {
+      switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "whisper_speech_feed_tts: no model_id provided\n");
+      return SWITCH_STATUS_FALSE;
+    }
+
+    /* format url*/
+    std::string url = "https://api.openai.com/v1/audio/speech";
+
+    /* create the JSON body */
+    cJSON * jResult = cJSON_CreateObject();
+    cJSON_AddStringToObject(jResult, "model", w->model_id);
+    cJSON_AddStringToObject(jResult, "input", text);
+    cJSON_AddStringToObject(jResult, "voice", w->voice_name);
+    cJSON_AddStringToObject(jResult, "response_format", "mp3");
+    if (w->speed) {
+      cJSON_AddStringToObject(jResult, "speed", w->speed);
+    }
+    char *json = cJSON_PrintUnformatted(jResult);
+
+    cJSON_Delete(jResult);
+
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "whisper_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();
+    w->conn = (void *) conn ;
+    conn->whisper = w;
+    conn->easy = easy;
+    conn->mh = mh;
+    conn->global = &global;
+    conn->hdr_list = NULL ;
+    conn->file = w->file;
+    conn->body = json;
+    conn->flushed = false;
+    
+
+    w->circularBuffer = (void *) new CircularBuffer_t(8192);
+
+    if (w->session_id) {
+      int err;
+      switch_codec_implementation_t read_impl;
+      switch_core_session_t *psession = switch_core_session_locate(w->session_id);
+      switch_core_session_get_read_impl(psession, &read_impl);
+      uint32_t samples_per_second = !strcasecmp(read_impl.iananame, "g722") ? read_impl.actual_samples_per_second : read_impl.samples_per_second;
+      if (mpg123_param(mh, MPG123_FORCE_RATE, samples_per_second /*Hz*/, 0) != MPG123_OK) {
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "Error mpg123_param!\n");
+        return SWITCH_STATUS_FALSE;
+      }
+    }
+
+    std::ostringstream api_key_stream;
+    api_key_stream << "Authorization: Bearer " << w->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());
+
+    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, "whisper_speech_feed_tts: called curl_multi_add_handle\n");
+
+
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+  switch_status_t whisper_speech_read_tts(whisper_t* w, void *data, size_t *datalen, switch_speech_flag_t *flags) {
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) w->circularBuffer;
+    std::vector<uint16_t> pcm_data;
+
+    {
+      switch_mutex_lock(w->mutex);
+      ConnInfo_t *conn = (ConnInfo_t *) w->conn;
+      if (w->response_code > 0 && w->response_code != 200) {
+        switch_mutex_unlock(w->mutex);
+        switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "whisper_speech_read_tts, returning failure\n") ;  
+        return SWITCH_STATUS_FALSE;
+      }
+      if (conn && conn->flushed) {
+        switch_mutex_unlock(w->mutex);
+        return SWITCH_STATUS_BREAK;
+      }
+      if (cBuffer->empty()) {
+        if (w->draining) {
+          switch_mutex_unlock(w->mutex);
+          return SWITCH_STATUS_BREAK;
+        }
+        /* no audio available yet so send silence */
+        memset(data, 255, *datalen);
+        switch_mutex_unlock(w->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(w->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 whisper_speech_flush_tts(whisper_t* w) {
+    bool download_complete = w->response_code == 200;
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "whisper_speech_flush_tts, download complete? %s\n", download_complete ? "yes" : "no") ;  
+
+    ConnInfo_t *conn = (ConnInfo_t *) w->conn;
+    CircularBuffer_t *cBuffer = (CircularBuffer_t *) w->circularBuffer;
+    delete cBuffer;
+    w->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", w->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 (w->cache_filename) {
+          switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_DEBUG, "removing audio cache file %s because download was interrupted\n", w->cache_filename);
+          if (unlink(w->cache_filename) != 0) {
+            switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_ERROR, "cleanupConn: error removing audio cache file %s: %d:%s\n", 
+              w->cache_filename, errno, strerror(errno));
+          }
+          free(w->cache_filename);
+          w->cache_filename = nullptr ;
+        }
+      }
+    }
+    if (w->session_id) {
+      switch_core_session_t* session = switch_core_session_locate(w->session_id);
+      if (session) {
+        switch_channel_t *channel = switch_core_session_get_channel(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_whisper_response_code", std::to_string(w->response_code).c_str());
+            if (w->cache_filename && w->response_code == 200) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_cache_filename", w->cache_filename);
+            }
+            if (w->response_code != 200 && w->err_msg) {
+              switch_event_add_header_string(event, SWITCH_STACK_BOTTOM, "variable_tts_error", w->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");
+        }
+        switch_core_session_rwunlock(session);
+      }
+    }
+    return SWITCH_STATUS_SUCCESS;
+  }
+
+	switch_status_t whisper_speech_close(whisper_t* w) {
+    switch_log_printf(SWITCH_CHANNEL_LOG, SWITCH_LOG_INFO, "whisper_speech_close\n") ;
+		return SWITCH_STATUS_SUCCESS;
+	}
+}

mod_whisper_tts/whisper_glue.h

@@ -0,0 +1,12 @@
+#ifndef __WHISPER_GLUE_H__
+#define __WHISPER_GLUE_H__
+
+switch_status_t whisper_speech_load();
+switch_status_t whisper_speech_open(whisper_t* whisper);
+switch_status_t whisper_speech_feed_tts(whisper_t* whisper, char* text, switch_speech_flag_t *flags);
+switch_status_t whisper_speech_read_tts(whisper_t* whisper, void *data, size_t *datalen, switch_speech_flag_t *flags);
+switch_status_t whisper_speech_flush_tts(whisper_t* whisper);
+switch_status_t whisper_speech_close(whisper_t* whisper);
+switch_status_t whisper_speech_unload();
+
+#endif