Feat/ambient sounds (#678)

* initial support for coaching mode in conference

* wip

* wip

* add support for answer verb

* wip

* wip

* wip

* wip

* wip

* updates to rename option to dub

* wip

* wip

* wip

* update verb-specs

* wip

* wip

* wip

* wip

* wip

* wip

* wip

* wip

* add option to boost audio signal in main channel

* wip

* wip

* wip

* wip

* wip

* wip

* for now, bypass use of streaming apis when generating tts audio for dub tracks

* add nested dub to dial

* wip

* add support for filler noise

* kill filler noise when gather killed

* wip

* wip

* while using sayOnTrack, we have to enclose the say command in double quotes

* disableTtsStreaming = false

* allow transcribe of b leg only on dial verb

* dub.say can either be text or object like say verb with text and synthesizer

* remove loop for sayOnTrack

* update speech-utils

* fixes for testing transcribe verb and support for dub and boostAudioSignal in lcc commands

* add dial.boostAudioSignal

* fix bug where session-level recognizer settings incorrectly overwrite verb-level settings

* update verb specs

* update dial to support array of dub verbs

* fix bug setting gain

* lint

* wip

* update speech-utils

* use new endpoint methods for mod_dub

---------

Co-authored-by: Dave Horton <daveh@beachdognet.com>

lib/tasks/transcribe.js

@@ -32,8 +32,22 @@ class TaskTranscribe extends SttTask {
     }
 
     /* for nested transcribe in dial, unless the app explicitly says so we want to transcribe both legs */
-    if (this.parentTask?.name === TaskName.Dial && this.separateRecognitionPerChannel !== false) {
-      this.separateRecognitionPerChannel = true;
+    if (this.parentTask?.name === TaskName.Dial) {
+      if (this.data.channel === 1 || this.data.channel === 2) {
+        /* transcribe only the channel specified */
+        this.separateRecognitionPerChannel = false;
+        this.channel = this.data.channel;
+        logger.debug(`TaskTranscribe: transcribing only channel ${this.channel} in the Dial verb`);
+      }
+      else if (this.separateRecognitionPerChannel !== false) {
+        this.separateRecognitionPerChannel = true;
+      }
+      else {
+        this.channel = 1;
+      }
+    }
+    else {
+      this.channel = 1;
     }
 
     this.childSpan = [null, null];
@@ -51,6 +65,14 @@ class TaskTranscribe extends SttTask {
 
   get name() { return TaskName.Transcribe; }
 
+  get transcribing1() {
+    return this.channel === 1 || this.separateRecognitionPerChannel;
+  }
+
+  get transcribing2() {
+    return this.channel === 2 || this.separateRecognitionPerChannel && this.ep2;
+  }
+
   async exec(cs, {ep, ep2}) {
     await super.exec(cs, {ep, ep2});
 
@@ -73,8 +95,10 @@ class TaskTranscribe extends SttTask {
     }
 
     try {
-      await this._startTranscribing(cs, ep, 1);
-      if (this.separateRecognitionPerChannel && ep2) {
+      if (this.transcribing1) {
+        await this._startTranscribing(cs, ep, 1);
+      }
+      if (this.transcribing2) {
         await this._startTranscribing(cs, ep2, 2);
       }
 
@@ -91,7 +115,7 @@ class TaskTranscribe extends SttTask {
 
   async _stopTranscription() {
     let stopTranscription = false;
-    if (this.ep?.connected) {
+    if (this.transcribing1 && this.ep?.connected) {
       stopTranscription = true;
       this.ep.stopTranscription({
         vendor: this.vendor,
@@ -99,7 +123,7 @@ class TaskTranscribe extends SttTask {
       })
         .catch((err) => this.logger.info(err, 'Error TaskTranscribe:kill'));
     }
-    if (this.separateRecognitionPerChannel && this.ep2 && this.ep2.connected) {
+    if (this.transcribing2 && this.ep2.connected) {
       stopTranscription = true;
       this.ep2.stopTranscription({vendor: this.vendor, bugname: this.bugname})
         .catch((err) => this.logger.info(err, 'Error TaskTranscribe:kill'));
@@ -128,10 +152,8 @@ class TaskTranscribe extends SttTask {
           break;
         case TranscribeStatus.Resume:
           this.paused = false;
-          await this._startTranscribing(this.cs, this.ep, 1);
-          if (this.separateRecognitionPerChannel && this.ep2) {
-            await this._startTranscribing(this.cs, this.ep2, 2);
-          }
+          if (this.transcribing1) await this._startTranscribing(this.cs, this.ep, 1);
+          if (this.transcribing2) await this._startTranscribing(this.cs, this.ep2, 2);
           break;
       }
     }
@@ -294,7 +316,7 @@ class TaskTranscribe extends SttTask {
       vendor: this.vendor,
       interim: this.interim ? true : false,
       locale: this.language,
-      channels: /*this.separateRecognitionPerChannel ? 2 : */ 1,
+      channels: 1,
       bugname: this.bugname,
       hostport: this.hostport
     });
@@ -303,12 +325,12 @@ class TaskTranscribe extends SttTask {
   async _onTranscription(cs, ep, channel, evt, fsEvent) {
     // make sure this is not a transcript from answering machine detection
     const bugname = fsEvent.getHeader('media-bugname');
+    const finished = fsEvent.getHeader('transcription-session-finished');
     if (bugname && this.bugname !== bugname) return;
     if (this.paused) {
       this.logger.debug({evt}, 'TaskTranscribe:_onTranscription - paused, ignoring transcript');
     }
 
-
     if (this.vendor === 'ibm' && evt?.state === 'listening') return;
 
     if (this.vendor === 'deepgram' && evt.type === 'UtteranceEnd') {
@@ -319,8 +341,9 @@ class TaskTranscribe extends SttTask {
       else {
         this.logger.debug('Gather:_onTranscription - got UtteranceEnd event from deepgram, return buffered transcript');
         evt = this.consolidateTranscripts(this._bufferedTranscripts, 1, this.language, this.vendor);
+        evt.is_final = true;
         this._bufferedTranscripts = [];
-        this._resolve('speech', evt);
+        this._resolve(channel, evt);
       }
       return;
     }
@@ -334,31 +357,87 @@ class TaskTranscribe extends SttTask {
       return;
     }
 
-    if (evt.alternatives[0]?.transcript === '' && !cs.callGone && !this.killed) {
-      if (['microsoft', 'deepgram'].includes(this.vendor)) {
-        this.logger.info({evt}, 'TaskTranscribe:_onTranscription - got empty transcript, continue listening');
+    let emptyTranscript = false;
+    if (evt.is_final) {
+      if (evt.alternatives[0].transcript === '' && !cs.callGone && !this.killed) {
+        emptyTranscript = true;
+        if (finished === 'true' &&
+          ['microsoft', 'deepgram'].includes(this.vendor) &&
+          this._bufferedTranscripts.length === 0) {
+          this.logger.debug({evt}, 'TaskGather:_onTranscription - got empty transcript from old gather, disregarding');
+          return;
+        }
+        else if (this.vendor !== 'deepgram') {
+          this.logger.info({evt}, 'TaskGather:_onTranscription - got empty transcript, continue listening');
+          return;
+        }
+        else if (this.isContinuousAsr) {
+          this.logger.info({evt},
+            'TaskGather:_onTranscription - got empty deepgram transcript during continous asr, continue listening');
+          return;
+        }
+        else if (this.vendor === 'deepgram' && this._bufferedTranscripts.length > 0) {
+          this.logger.info({evt},
+            'TaskGather:_onTranscription - got empty transcript from deepgram, return the buffered transcripts');
+        }
+      }
+      if (this.isContinuousAsr) {
+        /* append the transcript and start listening again for asrTimeout */
+        const t = evt.alternatives[0].transcript;
+        if (t) {
+          /* remove trailing punctuation */
+          if (/[,;:\.!\?]$/.test(t)) {
+            this.logger.debug('TaskGather:_onTranscription - removing trailing punctuation');
+            evt.alternatives[0].transcript = t.slice(0, -1);
+          }
+        }
+        this.logger.info({evt}, 'TaskGather:_onTranscription - got transcript during continous asr');
+        this._bufferedTranscripts.push(evt);
+        this._startAsrTimer(channel);
+
+        /* some STT engines will keep listening after a final response, so no need to restart */
+        if (!['soniox', 'aws', 'microsoft', 'deepgram'].includes(this.vendor)) this._startTranscribing(cs, ep, channel);
       }
       else {
-        this.logger.info({evt}, 'TaskTranscribe:_onTranscription - got empty transcript, listen again');
-        this._transcribe(ep);
-      }
-      return;
-    }
+        if (this.vendor === 'soniox') {
+          /* compile transcripts into one */
+          this._sonioxTranscripts.push(evt.vendor.finalWords);
+          evt = this.compileSonioxTranscripts(this._sonioxTranscripts, 1, this.language);
+          this._sonioxTranscripts = [];
+        }
+        else if (this.vendor === 'deepgram') {
+          /* compile transcripts into one */
+          if (!emptyTranscript) this._bufferedTranscripts.push(evt);
 
-    if (this.vendor === 'soniox') {
-      /* compile transcripts into one */
-      this._sonioxTranscripts.push(evt.vendor.finalWords);
-      if (evt.is_final) {
-        evt = this.compileSonioxTranscripts(this._sonioxTranscripts, 1, this.language);
-        this._sonioxTranscripts = [];
+          /* deepgram can send an empty and final transcript; only if we have any buffered should we resolve */
+          if (this._bufferedTranscripts.length === 0) return;
+          evt = this.consolidateTranscripts(this._bufferedTranscripts, channel, this.language);
+          this._bufferedTranscripts = [];
+        }
+
+        /* here is where we return a final transcript */
+        this.logger.debug({evt}, 'TaskTranscribe:_onTranscription - sending final transcript');
+        this._resolve(channel, evt);
+        /* some STT engines will keep listening after a final response, so no need to restart */
+        if (!['soniox', 'aws', 'microsoft', 'deepgram'].includes(this.vendor)) this._startTranscribing(cs, ep, channel);
       }
     }
+    else {
+      /* interim transcript */
 
-    if (this.isContinuousAsr && evt.is_final) {
-      this._bufferedTranscripts.push(evt);
-      this._startAsrTimer(channel);
-    } else {
-      await this._resolve(channel, evt);
+      /* deepgram can send a non-final transcript but with words that are final, so we need to buffer */
+      if (this.vendor === 'deepgram') {
+        const originalEvent = evt.vendor.evt;
+        if (originalEvent.is_final && evt.alternatives[0].transcript !== '') {
+          this.logger.debug({evt}, 'Gather:_onTranscription - buffering a completed (partial) deepgram transcript');
+          this._bufferedTranscripts.push(evt);
+        }
+      }
+
+      if (this.interim) {
+        this.logger.debug({evt}, 'TaskTranscribe:_onTranscription - sending interim transcript');
+        this._resolve(channel, evt);
+      }
     }
   }
 
@@ -479,6 +558,7 @@ class TaskTranscribe extends SttTask {
 
       if (this.vendor === 'nuance') {
         const {code, error} = evt;
+        //TODO: fix below, currently _resolve does not send timeout events
         if (code === 404 && error === 'No speech') return this._resolve('timeout');
         if (code === 413 && error === 'Too much speech') return this._resolve('timeout');
       }