feat: update speech-ultil version 1.0.1 (#275)

* feat: update speech-ultil version 1.0.1

* feat: update speech-ultil version 1.0.1

* more fixes for custom stt

* more fixes

* fixes

* update drachtio-fsmrf

* pass url to mod_jambonz_transcribe

* transcription utils: handle custom results

* handle custom speech vendor errors

* add support for hints to custom speech

* change to custom speech options

* send hints as an array for custom speech

* update latest speech-utils

* transcribe: changes to support soniox

* bugfix: soniox transcribe

---------

Co-authored-by: Quan HL <quanluuhoang8@gmail.com>
Co-authored-by: Dave Horton <daveh@beachdognet.com>

lib/utils/constants.json

@@ -110,6 +110,12 @@
     "NoSpeechDetected": "azure_transcribe::no_speech_detected",
     "VadDetected": "azure_transcribe::vad_detected"
   },
+  "JambonzTranscriptionEvents": {
+    "Transcription": "jambonz_transcribe::transcription",
+    "ConnectFailure": "jambonz_transcribe::connect_failed",
+    "Connect": "jambonz_transcribe::connect",
+    "Error": "jambonz_transcribe::error"
+  },
   "ListenEvents": {
     "Connect": "mod_audio_fork::connect",
     "ConnectFailure": "mod_audio_fork::connect_failed",

lib/utils/db-utils.js

@@ -66,6 +66,12 @@ const speechMapper = (cred) => {
       const o = JSON.parse(decrypt(credential));
       obj.api_key = o.api_key;
     }
+    else if (obj.vendor.startsWith('custom:')) {
+      const o = JSON.parse(decrypt(credential));
+      obj.auth_token = o.auth_token;
+      obj.custom_stt_url = o.custom_stt_url;
+      obj.custom_tts_url = o.custom_tts_url;
+    }
   } catch (err) {
     console.log(err);
   }
@@ -83,53 +89,13 @@ module.exports = (logger, srf) => {
     const [r2] = await pp.query(sqlSpeechCredentials, account_sid);
     const speech = r2.map(speechMapper);
 
-    /* search at the service provider level if we don't find it at the account level */
-    const haveGoogle = speech.find((s) => s.vendor === 'google');
-    const haveAws = speech.find((s) => s.vendor === 'aws');
-    const haveMicrosoft = speech.find((s) => s.vendor === 'microsoft');
-    const haveWellsaid = speech.find((s) => s.vendor === 'wellsaid');
-    const haveNuance = speech.find((s) => s.vendor === 'nuance');
-    const haveDeepgram = speech.find((s) => s.vendor === 'deepgram');
-    const haveSoniox = speech.find((s) => s.vendor === 'soniox');
-    const haveIbm = speech.find((s) => s.vendor === 'ibm');
-    if (!haveGoogle || !haveAws || !haveMicrosoft || !haveWellsaid ||
-      !haveNuance || !haveIbm || !haveDeepgram || !haveSoniox) {
-      const [r3] = await pp.query(sqlSpeechCredentialsForSP, account_sid);
-      if (r3.length) {
-        if (!haveGoogle) {
-          const google = r3.find((s) => s.vendor === 'google');
-          if (google) speech.push(speechMapper(google));
-        }
-        if (!haveAws) {
-          const aws = r3.find((s) => s.vendor === 'aws');
-          if (aws) speech.push(speechMapper(aws));
-        }
-        if (!haveMicrosoft) {
-          const ms = r3.find((s) => s.vendor === 'microsoft');
-          if (ms) speech.push(speechMapper(ms));
-        }
-        if (!haveWellsaid) {
-          const wellsaid = r3.find((s) => s.vendor === 'wellsaid');
-          if (wellsaid) speech.push(speechMapper(wellsaid));
-        }
-        if (!haveNuance) {
-          const nuance = r3.find((s) => s.vendor === 'nuance');
-          if (nuance) speech.push(speechMapper(nuance));
-        }
-        if (!haveDeepgram) {
-          const deepgram = r3.find((s) => s.vendor === 'deepgram');
-          if (deepgram) speech.push(speechMapper(deepgram));
-        }
-        if (!haveSoniox) {
-          const soniox = r3.find((s) => s.vendor === 'soniox');
-          if (soniox) speech.push(speechMapper(soniox));
-        }
-        if (!haveIbm) {
-          const ibm = r3.find((s) => s.vendor === 'ibm');
-          if (ibm) speech.push(speechMapper(ibm));
-        }
+    /* add service provider creds unless we have that vendor at the account level */
+    const [r3] = await pp.query(sqlSpeechCredentialsForSP, account_sid);
+    r3.forEach((s) => {
+      if (!speech.find((s2) => s2.vendor === s.vendor)) {
+        speech.push(speechMapper(s));
       }
-    }
+    });
 
     return {
       ...r[0],

lib/utils/install-srf-locals.js

@@ -138,7 +138,6 @@ function installSrfLocals(srf, logger) {
     retrieveCall,
     listCalls,
     deleteCall,
-    synthAudio,
     createHash,
     retrieveHash,
     deleteKey,
@@ -151,11 +150,17 @@ function installSrfLocals(srf, logger) {
     pushBack,
     popFront,
     removeFromList,
-    lengthOfList,
     getListPosition,
+    lengthOfList,
+  } = require('@jambonz/realtimedb-helpers')({
+    host: process.env.JAMBONES_REDIS_HOST,
+    port: process.env.JAMBONES_REDIS_PORT || 6379
+  }, logger, tracer);
+  const {
+    synthAudio,
     getNuanceAccessToken,
     getIbmAccessToken,
-  } = require('@jambonz/realtimedb-helpers')({
+  } = require('@jambonz/speech-utils')({
     host: process.env.JAMBONES_REDIS_HOST,
     port: process.env.JAMBONES_REDIS_PORT || 6379
   }, logger, tracer);

lib/utils/transcription-utils.js

@@ -6,7 +6,8 @@ const {
   NuanceTranscriptionEvents,
   DeepgramTranscriptionEvents,
   SonioxTranscriptionEvents,
-  NvidiaTranscriptionEvents
+  NvidiaTranscriptionEvents,
+  JambonzTranscriptionEvents
 } = require('./constants');
 
 const stickyVars = {
@@ -223,6 +224,15 @@ const normalizeGoogle = (evt, channel, language) => {
   };
 };
 
+const normalizeCustom = (evt, channel, language) => {
+  return {
+    language_code: language,
+    channel_tag: channel,
+    is_final: evt.is_final,
+    alternatives: [evt.alternatives[0]]
+  };
+};
+
 const normalizeNuance = (evt, channel, language) => {
   const copy = JSON.parse(JSON.stringify(evt));
   return {
@@ -302,6 +312,9 @@ module.exports = (logger) => {
       case 'soniox':
         return normalizeSoniox(evt, channel, language);
       default:
+        if (vendor.startsWith('custom:')) {
+          return normalizeCustom(evt, channel, language);
+        }
         logger.error(`Unknown vendor ${vendor}`);
         return evt;
     }
@@ -311,6 +324,7 @@ module.exports = (logger) => {
     let opts = {};
     const {enable, voiceMs = 0, mode = -1} = rOpts.vad || {};
     const vad = {enable, voiceMs, mode};
+    const vendor = rOpts.vendor;
 
     /* voice activity detection works across vendors */
     opts = {
@@ -320,7 +334,7 @@ module.exports = (logger) => {
       ...(vad.enable && typeof vad.mode === 'number' && {RECOGNIZER_VAD_MODE: vad.mode}),
     };
 
-    if ('google' === rOpts.vendor) {
+    if ('google' === vendor) {
       opts = {
         ...opts,
         ...(sttCredentials &&
@@ -372,7 +386,7 @@ module.exports = (logger) => {
           {GOOGLE_SPEECH_METADATA_INDUSTRY_NAICS_CODE: rOpts.naicsCode}),
       };
     }
-    else if (['aws', 'polly'].includes(rOpts.vendor)) {
+    else if (['aws', 'polly'].includes(vendor)) {
       opts = {
         ...opts,
         ...(rOpts.vocabularyName && {AWS_VOCABULARY_NAME: rOpts.vocabularyName}),
@@ -385,7 +399,7 @@ module.exports = (logger) => {
         }),
       };
     }
-    else if ('microsoft' === rOpts.vendor) {
+    else if ('microsoft' === vendor) {
       opts = {
         ...opts,
         ...(rOpts.hints.length > 0 && typeof rOpts.hints[0] === 'string' &&
@@ -410,7 +424,7 @@ module.exports = (logger) => {
           {AZURE_SERVICE_ENDPOINT_ID: sttCredentials.custom_stt_endpoint})
       };
     }
-    else if ('nuance' === rOpts.vendor) {
+    else if ('nuance' === vendor) {
       /**
        * Note: all nuance options are in recognizer.nuanceOptions, should migrate
        * other vendor settings to similar nested structure
@@ -461,7 +475,7 @@ module.exports = (logger) => {
           {NUANCE_RESOURCES: JSON.stringify(nuanceOptions.resources)},
       };
     }
-    else if ('deepgram' === rOpts.vendor) {
+    else if ('deepgram' === vendor) {
       const {deepgramOptions = {}} = rOpts;
       opts = {
         ...opts,
@@ -505,7 +519,7 @@ module.exports = (logger) => {
           {DEEPGRAM_SPEECH_TAG: deepgramOptions.tag}
       };
     }
-    else if ('soniox' === rOpts.vendor) {
+    else if ('soniox' === vendor) {
       const {sonioxOptions = {}} = rOpts;
       const {storage = {}} = sonioxOptions;
       opts = {
@@ -528,7 +542,7 @@ module.exports = (logger) => {
         ...(storage?.id && storage?.disableSearch && {SONIOX_STORAGE_DISABLE_SEARCH: 1})
       };
     }
-    else if ('ibm' === rOpts.vendor) {
+    else if ('ibm' === vendor) {
       const {ibmOptions = {}} = rOpts;
       opts = {
         ...opts,
@@ -552,7 +566,7 @@ module.exports = (logger) => {
           {IBM_SPEECH_WATSON_LEARNING_OPT_OUT: ibmOptions.watsonLearningOptOut}
       };
     }
-    else if ('nvidia' === rOpts.vendor) {
+    else if ('nvidia' === vendor) {
       const {nvidiaOptions = {}} = rOpts;
       opts = {
         ...opts,
@@ -581,11 +595,29 @@ module.exports = (logger) => {
           {NVIDIA_CUSTOM_CONFIGURATION: JSON.stringify(nvidiaOptions.customConfiguration)}),
       };
     }
+    else if (vendor.startsWith('custom:')) {
+      let {options = {}} = rOpts;
+      const {auth_token, custom_stt_url} = sttCredentials;
+      options = {
+        ...options,
+        ...(rOpts.hints.length > 0 && typeof rOpts.hints[0] === 'string' &&
+        {hints: rOpts.hints}),
+        ...(rOpts.hints.length > 0 && typeof rOpts.hints[0] === 'object' &&
+        {hints: JSON.stringify(rOpts.hints)}),
+        ...(typeof rOpts.hintsBoost === 'number' && {hintsBoost: rOpts.hintsBoost})
+      };
 
-    stickyVars[rOpts.vendor].forEach((key) => {
+      opts = {
+        ...opts,
+        JAMBONZ_STT_API_KEY: auth_token,
+        JAMBONZ_STT_URL: custom_stt_url,
+        ...(Object.keys(options).length > 0 && {JAMBONZ_STT_OPTIONS: JSON.stringify(options)}),
+      };
+    }
+
+    (stickyVars[vendor] || []).forEach((key) => {
       if (!opts[key]) opts[key] = '';
     });
-    //logger.debug({opts}, 'recognizer channel vars');
     return opts;
   };
 
@@ -604,7 +636,6 @@ module.exports = (logger) => {
     ep.removeCustomEventListener(NuanceTranscriptionEvents.Transcription);
     ep.removeCustomEventListener(NuanceTranscriptionEvents.TranscriptionComplete);
     ep.removeCustomEventListener(NuanceTranscriptionEvents.StartOfSpeech);
-    ep.removeCustomEventListener(NuanceTranscriptionEvents.Error);
     ep.removeCustomEventListener(NuanceTranscriptionEvents.VadDetected);
 
     ep.removeCustomEventListener(DeepgramTranscriptionEvents.Transcription);
@@ -612,13 +643,17 @@ module.exports = (logger) => {
     ep.removeCustomEventListener(DeepgramTranscriptionEvents.ConnectFailure);
 
     ep.removeCustomEventListener(SonioxTranscriptionEvents.Transcription);
-    ep.removeCustomEventListener(SonioxTranscriptionEvents.Error);
 
     ep.removeCustomEventListener(NvidiaTranscriptionEvents.Transcription);
     ep.removeCustomEventListener(NvidiaTranscriptionEvents.TranscriptionComplete);
     ep.removeCustomEventListener(NvidiaTranscriptionEvents.StartOfSpeech);
-    ep.removeCustomEventListener(NvidiaTranscriptionEvents.Error);
     ep.removeCustomEventListener(NvidiaTranscriptionEvents.VadDetected);
+
+    ep.removeCustomEventListener(JambonzTranscriptionEvents.Transcription);
+    ep.removeCustomEventListener(JambonzTranscriptionEvents.Connect);
+    ep.removeCustomEventListener(JambonzTranscriptionEvents.ConnectFailure);
+
+    ep.removeCustomEventListener(JambonzTranscriptionEvents.Error);
   };
 
   const setSpeechCredentialsAtRuntime = (recognizer) => {