Feature/precache audio (#609)

* wip

* fix for establishing vendor etc

* more fixes

* avoid a pre-caching attempt if synth settings change

lib/config.js

@@ -120,6 +120,7 @@ const HTTP_TIMEOUT = 10000;
 const HTTP_PROXY_IP = process.env.JAMBONES_HTTP_PROXY_IP;
 const HTTP_PROXY_PORT = process.env.JAMBONES_HTTP_PROXY_PORT;
 const HTTP_PROXY_PROTOCOL = process.env.JAMBONES_HTTP_PROXY_PROTOCOL || 'http';
+const HTTP_USER_AGENT_HEADER = process.env.JAMBONES_HTTP_USER_AGENT_HEADER || 'jambonz';
 
 const OPTIONS_PING_INTERVAL = parseInt(process.env.OPTIONS_PING_INTERVAL, 10) || 30000;
 
@@ -129,6 +130,8 @@ const JAMBONZ_RECORD_WS_PASSWORD = process.env.JAMBONZ_RECORD_WS_PASSWORD || pro
 const JAMBONZ_DISABLE_DIAL_PAI_HEADER = process.env.JAMBONZ_DISABLE_DIAL_PAI_HEADER || false;
 const JAMBONES_DISABLE_DIRECT_P2P_CALL = process.env.JAMBONES_DISABLE_DIRECT_P2P_CALL || false;
 
+const JAMBONES_EAGERLY_PRE_CACHE_AUDIO = process.env.JAMBONES_EAGERLY_PRE_CACHE_AUDIO;
+
 module.exports = {
   JAMBONES_MYSQL_HOST,
   JAMBONES_MYSQL_USER,
@@ -151,6 +154,7 @@ module.exports = {
   JAMBONES_API_BASE_URL,
   JAMBONES_TIME_SERIES_HOST,
   JAMBONES_INJECT_CONTENT,
+  JAMBONES_EAGERLY_PRE_CACHE_AUDIO,
   JAMBONES_ESL_LISTEN_ADDRESS,
   JAMBONES_SBCS,
   JAMBONES_OTEL_ENABLED,
@@ -193,6 +197,7 @@ module.exports = {
   HTTP_PROXY_IP,
   HTTP_PROXY_PORT,
   HTTP_PROXY_PROTOCOL,
+  HTTP_USER_AGENT_HEADER,
   OPTIONS_PING_INTERVAL,
   RESPONSE_TIMEOUT_MS,
   JAMBONES_WS_HANDSHAKE_TIMEOUT_MS,

lib/middleware.js

@@ -382,7 +382,7 @@ module.exports = function(srf, logger) {
           },
           recognizer: {
             vendor: app.speech_recognizer_vendor,
-            ...(app.speech_synthesis_label && {label: app.speech_synthesis_label}),
+            ...(app.speech_recognizer_label && {label: app.speech_recognizer_label}),
             language: app.speech_recognizer_language,
             ...(app.fallback_speech_recognizer_vendor && {fallback_vendor: app.fallback_speech_recognizer_vendor}),
             ...(app.fallback_speech_recognizer_label && {fallback_label: app.fallback_speech_recognizer_label}),

lib/session/call-session.js

@@ -19,6 +19,7 @@ const HttpRequestor = require('../utils/http-requestor');
 const WsRequestor = require('../utils/ws-requestor');
 const {
   JAMBONES_INJECT_CONTENT,
+  JAMBONES_EAGERLY_PRE_CACHE_AUDIO,
   AWS_REGION,
 } = require('../config');
 const BackgroundTaskManager = require('../utils/background-task-manager');
@@ -1330,6 +1331,35 @@ Duration=${duration} `
     this.taskIdx = 0;
   }
 
+  _preCacheAudio(newTasks) {
+    for (const task of newTasks) {
+      if (task.name === TaskName.Config && task.hasSynthesizer) {
+        /* if they change synthesizer settings don't try to precache */
+        break;
+      }
+      if (task.name === TaskName.Say) {
+        /* identify vendor language, voice, and label */
+        const vendor = task.synthesizer.vendor && task.synthesizer.vendor !== 'default' ?
+          task.synthesizer.vendor :
+          this.speechSynthesisVendor;
+        const language = task.synthesizer.language && task.synthesizer.language !== 'default' ?
+          task.synthesizer.language :
+          this.speechSynthesisLanguage ;
+        const voice =  task.synthesizer.voice && task.synthesizer.voice !== 'default' ?
+          task.synthesizer.voice :
+          this.speechSynthesisVoice;
+        const label = task.synthesizer.label && task.synthesizer.label !== 'default' ?
+          task.synthesizer.label :
+          this.speechSynthesisLabel;
+
+        this.logger.info({vendor, language, voice, label},
+          'CallSession:_preCacheAudio - precaching audio for future prompt');
+        task._synthesizeWithSpecificVendor(this, this.ep, {vendor, language, voice, label, preCache: true})
+          .catch((err) => this.logger.error(err, 'CallSession:_preCacheAudio - error precaching audio'));
+      }
+    }
+  }
+
   /**
    * Append tasks to the current execution stack UNLESS there is a gather in the stack.
    * in that case, insert the tasks before the gather AND if the tasks include
@@ -1387,10 +1417,12 @@ Duration=${duration} `
             this.replaceApplication(t);
           }
           else if (JAMBONES_INJECT_CONTENT) {
+            if (JAMBONES_EAGERLY_PRE_CACHE_AUDIO) this._preCacheAudio(t);
             this._injectTasks(t);
             this.logger.info({tasks: listTaskNames(this.tasks)}, 'CallSession:_onCommand - updated task list');
           }
           else {
+            if (JAMBONES_EAGERLY_PRE_CACHE_AUDIO) this._preCacheAudio(t);
             this.tasks.push(...t);
             this.logger.info({tasks: listTaskNames(this.tasks)}, 'CallSession:_onCommand - updated task list');
           }

lib/tasks/say.js

@@ -59,7 +59,7 @@ class TaskSay extends Task {
     }
   }
 
-  async _synthesizeWithSpecificVendor(cs, ep, {vendor, language, voice, label}) {
+  async _synthesizeWithSpecificVendor(cs, ep, {vendor, language, voice, label, preCache = false}) {
     const {srf} = cs;
     const {updateSpeechCredentialLastUsed} = require('../utils/db-utils')(this.logger, srf);
     const {writeAlerts, AlertType, stats} = srf.locals;
@@ -97,7 +97,7 @@ class TaskSay extends Task {
       voice = this.options.voice_id || voice;
     }
 
-    this.logger.info({vendor, language, voice, model}, 'TaskSay:exec');
+    if (!preCache) this.logger.info({vendor, language, voice, model}, 'TaskSay:exec');
     try {
       if (!credentials) {
         writeAlerts({
@@ -120,11 +120,15 @@ class TaskSay extends Task {
         if (text.startsWith('silence_stream://')) return text;
 
         /* otel: trace time for tts */
-        const {span} = this.startChildSpan('tts-generation', {
-          'tts.vendor': vendor,
-          'tts.language': language,
-          'tts.voice': voice
-        });
+        let otelSpan;
+        if (!preCache)  {
+          const {span} = this.startChildSpan('tts-generation', {
+            'tts.vendor': vendor,
+            'tts.language': language,
+            'tts.voice': voice
+          });
+          otelSpan = span;
+        }
         try {
           const {filePath, servedFromCache, rtt} = await synthAudio(stats, {
             account_sid: cs.accountSid,
@@ -146,9 +150,9 @@ class TaskSay extends Task {
             updateSpeechCredentialLastUsed(credentials.speech_credential_sid)
               .catch(() => {/*already logged error */});
           }
-          span.setAttributes({'tts.cached': servedFromCache});
-          span.end();
-          if (!servedFromCache && rtt) {
+          if (otelSpan) otelSpan.setAttributes({'tts.cached': servedFromCache});
+          if (otelSpan) otelSpan.end();
+          if (!servedFromCache && rtt && !preCache) {
             this.notifyStatus({
               event: 'synthesized-audio',
               vendor,
@@ -160,7 +164,7 @@ class TaskSay extends Task {
           return filePath;
         } catch (err) {
           this.logger.info({err}, 'Error synthesizing tts');
-          span.end();
+          if (otelSpan) otelSpan.end();
           writeAlerts({
             account_sid: cs.accountSid,
             alert_type: AlertType.TTS_FAILURE,

lib/tasks/transcribe.js

@@ -41,7 +41,6 @@ class TaskTranscribe extends SttTask {
     /* buffer speech for continuous asr */
     this._bufferedTranscripts = [];
     this.bugname_prefix = 'transcribe_';
-    this.bugnames = new Map();
   }
 
   get name() { return TaskName.Transcribe; }
@@ -90,13 +89,13 @@ class TaskTranscribe extends SttTask {
       stopTranscription = true;
       this.ep.stopTranscription({
         vendor: this.vendor,
-        bugname: this.bugnames.get(1)
+        bugname: this.bugname
       })
         .catch((err) => this.logger.info(err, 'Error TaskTranscribe:kill'));
     }
     if (this.separateRecognitionPerChannel && this.ep2 && this.ep2.connected) {
       stopTranscription = true;
-      this.ep2.stopTranscription({vendor: this.vendor, bugname: this.bugnames.get(2)})
+      this.ep2.stopTranscription({vendor: this.vendor})
         .catch((err) => this.logger.info(err, 'Error TaskTranscribe:kill'));
     }
 
@@ -133,7 +132,6 @@ class TaskTranscribe extends SttTask {
   async _setSpeechHandlers(cs, ep, channel) {
     if (this[`_speechHandlersSet_${channel}`]) return;
     this[`_speechHandlersSet_${channel}`] = true;
-    let bugname;
 
     /* some special deepgram logic */
     if (this.vendor === 'deepgram') {
@@ -143,7 +141,7 @@ class TaskTranscribe extends SttTask {
     const opts = this.setChannelVarsForStt(this, this.sttCredentials, this.data.recognizer);
     switch (this.vendor) {
       case 'google':
-        bugname = `${this.bugname_prefix}google_transcribe`;
+        this.bugname = `${this.bugname_prefix}google_transcribe`;
         this.addCustomEventListener(ep, GoogleTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         this.addCustomEventListener(ep, GoogleTranscriptionEvents.NoAudioDetected,
@@ -154,7 +152,7 @@ class TaskTranscribe extends SttTask {
 
       case 'aws':
       case 'polly':
-        bugname = `${this.bugname_prefix}aws_transcribe`;
+        this.bugname = `${this.bugname_prefix}aws_transcribe`;
         this.addCustomEventListener(ep, AwsTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         this.addCustomEventListener(ep, AwsTranscriptionEvents.NoAudioDetected,
@@ -163,19 +161,19 @@ class TaskTranscribe extends SttTask {
           this._onMaxDurationExceeded.bind(this, cs, ep, channel));
         break;
       case 'microsoft':
-        bugname = `${this.bugname_prefix}azure_transcribe`;
+        this.bugname = `${this.bugname_prefix}azure_transcribe`;
         this.addCustomEventListener(ep, AzureTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         this.addCustomEventListener(ep, AzureTranscriptionEvents.NoSpeechDetected,
           this._onNoAudio.bind(this, cs, ep, channel));
         break;
       case 'nuance':
-        bugname = `${this.bugname_prefix}nuance_transcribe`;
+        this.bugname = `${this.bugname_prefix}nuance_transcribe`;
         this.addCustomEventListener(ep, NuanceTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         break;
       case 'deepgram':
-        bugname = `${this.bugname_prefix}deepgram_transcribe`;
+        this.bugname = `${this.bugname_prefix}deepgram_transcribe`;
         this.addCustomEventListener(ep, DeepgramTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         this.addCustomEventListener(ep, DeepgramTranscriptionEvents.Connect,
@@ -188,12 +186,12 @@ class TaskTranscribe extends SttTask {
 
         break;
       case 'soniox':
-        bugname = `${this.bugname_prefix}soniox_transcribe`;
+        this.bugname = `${this.bugname_prefix}soniox_transcribe`;
         this.addCustomEventListener(ep, SonioxTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         break;
       case 'cobalt':
-        bugname = `${this.bugname_prefix}cobalt_transcribe`;
+        this.bugname = `${this.bugname_prefix}cobalt_transcribe`;
         this.addCustomEventListener(ep, CobaltTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
 
@@ -223,7 +221,7 @@ class TaskTranscribe extends SttTask {
         break;
 
       case 'ibm':
-        bugname = `${this.bugname_prefix}ibm_transcribe`;
+        this.bugname = `${this.bugname_prefix}ibm_transcribe`;
         this.addCustomEventListener(ep, IbmTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         this.addCustomEventListener(ep, IbmTranscriptionEvents.Connect,
@@ -233,13 +231,13 @@ class TaskTranscribe extends SttTask {
         break;
 
       case 'nvidia':
-        bugname = `${this.bugname_prefix}nvidia_transcribe`;
+        this.bugname = `${this.bugname_prefix}nvidia_transcribe`;
         this.addCustomEventListener(ep, NvidiaTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         break;
 
       case 'assemblyai':
-        bugname = `${this.bugname_prefix}assemblyai_transcribe`;
+        this.bugname = `${this.bugname_prefix}assemblyai_transcribe`;
         this.addCustomEventListener(ep, AssemblyAiTranscriptionEvents.Transcription,
           this._onTranscription.bind(this, cs, ep, channel));
         this.addCustomEventListener(ep,
@@ -251,7 +249,7 @@ class TaskTranscribe extends SttTask {
 
       default:
         if (this.vendor.startsWith('custom:')) {
-          bugname = `${this.bugname_prefix}${this.vendor}_transcribe`;
+          this.bugname = `${this.bugname_prefix}${this.vendor}_transcribe`;
           this.addCustomEventListener(ep, JambonzTranscriptionEvents.Transcription,
             this._onTranscription.bind(this, cs, ep, channel));
           this.addCustomEventListener(ep, JambonzTranscriptionEvents.Connect, this._onVendorConnect.bind(this, cs, ep));
@@ -265,33 +263,31 @@ class TaskTranscribe extends SttTask {
           throw new Error(`Invalid vendor ${this.vendor}`);
         }
     }
-    // save dedicated bugname for each endpoint
-    this.bugnames.set(channel, `${bugname}_${Date.now()}`);
+
     /* common handler for all stt engine errors */
-    this.addCustomEventListener(ep, JambonzTranscriptionEvents.Error, this._onJambonzError.bind(this, cs, ep, channel));
+    this.addCustomEventListener(ep, JambonzTranscriptionEvents.Error, this._onJambonzError.bind(this, cs, ep));
     await ep.set(opts)
       .catch((err) => this.logger.info(err, 'Error setting channel variables'));
   }
 
   async _startTranscribing(cs, ep, channel) {
     await this._setSpeechHandlers(cs, ep, channel);
-    await this._transcribe(ep, channel);
+    await this._transcribe(ep);
 
     /* start child span for this channel */
     const {span, ctx} = this.startChildSpan(`${STT_LISTEN_SPAN_NAME}:${channel}`);
     this.childSpan[channel - 1] = {span, ctx};
   }
 
-  async _transcribe(ep, channel) {
+  async _transcribe(ep) {
     this.logger.debug(
-      `TaskTranscribe:_transcribe - starting transcription vendor
- ${this.vendor} bugname ${this.bugnames.get(channel)}`);
+      `TaskTranscribe:_transcribe - starting transcription vendor ${this.vendor} bugname ${this.bugname}`);
     await ep.startTranscription({
       vendor: this.vendor,
       interim: this.interim ? true : false,
       locale: this.language,
       channels: /*this.separateRecognitionPerChannel ? 2 : */ 1,
-      bugname: this.bugnames.get(channel),
+      bugname: this.bugname,
       hostport: this.hostport
     });
   }
@@ -299,7 +295,7 @@ 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');
-    if (bugname && this.bugnames.get(channel) !== bugname) return;
+    if (bugname && this.bugname !== bugname) return;
 
     if (this.vendor === 'ibm' && evt?.state === 'listening') return;
 
@@ -332,7 +328,7 @@ class TaskTranscribe extends SttTask {
       }
       else {
         this.logger.info({evt}, 'TaskTranscribe:_onTranscription - got empty transcript, listen again');
-        this._transcribe(ep, channel);
+        this._transcribe(ep);
       }
       return;
     }
@@ -411,7 +407,7 @@ class TaskTranscribe extends SttTask {
       });
       this.childSpan[channel - 1].span.end();
     }
-    this._transcribe(ep, channel);
+    this._transcribe(ep);
 
     /* start new child span for this channel */
     const {span, ctx} = this.startChildSpan(`${STT_LISTEN_SPAN_NAME}:${channel}`);
@@ -428,7 +424,7 @@ class TaskTranscribe extends SttTask {
       this.childSpan[channel - 1].span.end();
     }
 
-    this._transcribe(ep, channel);
+    this._transcribe(ep);
 
     /* start new child span for this channel */
     const {span, ctx} = this.startChildSpan(`${STT_LISTEN_SPAN_NAME}:${channel}`);
@@ -442,12 +438,12 @@ class TaskTranscribe extends SttTask {
     }
   }
 
-  async _onJambonzError(cs, _ep, channel, evt) {
+  async _onJambonzError(cs, _ep, evt) {
     this.logger.info({evt}, 'TaskTranscribe:_onJambonzError');
     if (this.isHandledByPrimaryProvider && this.fallbackVendor) {
       _ep.stopTranscription({
         vendor: this.vendor,
-        bugname: this.bugnames.get(channel)
+        bugname: this.bugname
       })
         .catch((err) => this.logger.error({err}, `Error stopping transcription for primary vendor ${this.vendor}`));
       const {updateSpeechCredentialLastUsed} = require('../utils/db-utils')(this.logger, cs.srf);

lib/utils/http-requestor.js

@@ -14,6 +14,7 @@ const {
   HTTP_PROXY_PORT,
   HTTP_PROXY_PROTOCOL,
   NODE_ENV,
+  HTTP_USER_AGENT_HEADER,
 } = require('../config');
 
 const toBase64 = (str) => Buffer.from(str || '', 'utf8').toString('base64');
@@ -116,6 +117,10 @@ class HttpRequestor extends BaseRequestor {
     const url = hook.url || hook;
     const method = hook.method || 'POST';
     let buf = '';
+    httpHeaders = {
+      ...httpHeaders,
+      ...(HTTP_USER_AGENT_HEADER && {'user-agent' : HTTP_USER_AGENT_HEADER})
+    };
 
     assert.ok(url, 'HttpRequestor:request url was not provided');
     assert.ok, (['GET', 'POST'].includes(method), `HttpRequestor:request method must be 'GET' or 'POST' not ${method}`);

lib/utils/ws-requestor.js

@@ -9,7 +9,8 @@ const {
   JAMBONES_WS_PING_INTERVAL_MS,
   MAX_RECONNECTS,
   JAMBONES_WS_HANDSHAKE_TIMEOUT_MS,
-  JAMBONES_WS_MAX_PAYLOAD
+  JAMBONES_WS_MAX_PAYLOAD,
+  HTTP_USER_AGENT_HEADER
 } = require('../config');
 
 class WsRequestor extends BaseRequestor {
@@ -228,6 +229,9 @@ class WsRequestor extends BaseRequestor {
         maxRedirects: 2,
         handshakeTimeout,
         maxPayload: JAMBONES_WS_MAX_PAYLOAD ? parseInt(JAMBONES_WS_MAX_PAYLOAD) : 24 * 1024,
+        headers: {
+          ...(HTTP_USER_AGENT_HEADER && {'user-agent' : HTTP_USER_AGENT_HEADER})
+        }
       };
       if (this.username && this.password) opts = {...opts, auth: `${this.username}:${this.password}`};
 

package-lock.json

@@ -7530,9 +7530,9 @@
       "dev": true
     },
     "node_modules/follow-redirects": {
-      "version": "1.15.2",
-      "resolved": "https://registry.npmjs.org/follow-redirects/-/follow-redirects-1.15.2.tgz",
-      "integrity": "sha512-VQLG33o04KaQ8uYi2tVNbdrWp1QWxNNea+nmIB4EVM28v0hmP17z7aG1+wAkNzVq4KeXTq3221ye5qTJP91JwA==",
+      "version": "1.15.4",
+      "resolved": "https://registry.npmjs.org/follow-redirects/-/follow-redirects-1.15.4.tgz",
+      "integrity": "sha512-Cr4D/5wlrb0z9dgERpUL3LrmPKVDsETIJhaCMeDfuFYcqa5bldGV6wBsAN6X/vxlXQtFBMrXdXxdL8CbDTGniw==",
       "funding": [
         {
           "type": "individual",
@@ -18056,9 +18056,9 @@
       "dev": true
     },
     "follow-redirects": {
-      "version": "1.15.2",
-      "resolved": "https://registry.npmjs.org/follow-redirects/-/follow-redirects-1.15.2.tgz",
-      "integrity": "sha512-VQLG33o04KaQ8uYi2tVNbdrWp1QWxNNea+nmIB4EVM28v0hmP17z7aG1+wAkNzVq4KeXTq3221ye5qTJP91JwA=="
+      "version": "1.15.4",
+      "resolved": "https://registry.npmjs.org/follow-redirects/-/follow-redirects-1.15.4.tgz",
+      "integrity": "sha512-Cr4D/5wlrb0z9dgERpUL3LrmPKVDsETIJhaCMeDfuFYcqa5bldGV6wBsAN6X/vxlXQtFBMrXdXxdL8CbDTGniw=="
     },
     "for-each": {
       "version": "0.3.3",

test/create-call-test.js

@@ -99,6 +99,8 @@ test('test create-call call-hook basic authentication', async(t) => {
     let obj = await getJSON(`http:127.0.0.1:3100/lastRequest/${from}`)
     t.ok(obj.headers.Authorization = 'Basic dXNlcm5hbWU6cGFzc3dvcmQ=',
       'create-call: call-hook contains basic authentication header');
+    t.ok(obj.headers['user-agent'] = 'jambonz',
+    'create-call: call-hook contains user-agent header');
     disconnect();
   } catch (err) {
     console.log(`error received: ${err}`);