ben.carrasco/jambonz-feature-server
1
0
Fork 0
mirror of https://github.com/jambonz/jambonz-feature-server.git synced 2025-12-20 16:50:39 +00:00
Code Issues Packages Projects Releases Wiki Activity

Feature/nuance stt (#185)

Browse Source 
* initial changes to gather to support nuance stt

* updateSpeechCredentialLastUsed could be called without a speech_credential_sid if credentials are passed in the flow

* fix bugname

* typo

* added handlers for nuance

* logging

* major refactor of parsing transcriptions

* initial support for nuance in transcribe verb

* updates from testing

* cleanup some tests

* update action

* typo

* gather: start nuance timers after say/play completes

* update drachtio-fsrmf

* refactor some code

* typo

* log nuance error detail

* timeout handling

* typo

* handle nuance 413 response when recognition times out

* typo in specs.json

* add support for nuance resources

* fixes and tests for transcribe

* remove logging from test

* initial support for kryptonEndpoint

* try getting access token even when using krypton

* typo in kryptonEndpoint property

* add support for Nuance tts

* parse nuance voice and model for tts

* use nuance credentials from db

* update to db-helpers@0.7.0 with caching option

* add support for azure audio logging in gather/transcribe

* sync package-lock.json
This commit is contained in:
Dave Horton
2022-11-01 12:23:49 -04:00
committed by GitHub
parent 203b9774ca
commit 509bb065bb
20 changed files with 1100 additions and 974 deletions
Download Patch File Download Diff File Expand all files Collapse all files

302
lib/tasks/gather.js
View File

@@ -3,25 +3,22 @@ const {
TaskName,
TaskPreconditions,
GoogleTranscriptionEvents,
NuanceTranscriptionEvents,
AwsTranscriptionEvents,
AzureTranscriptionEvents
} = require('../utils/constants');
const makeTask = require('./make_task');
const assert = require('assert');
//const GATHER_STABILITY_THRESHOLD = Number(process.env.JAMBONZ_GATHER_STABILITY_THRESHOLD || 0.7);
const compileTranscripts = (logger, evt, arr) => {
//logger.debug({arr, evt}, 'compile transcripts');
if (!Array.isArray(arr) || arr.length === 0) return;
let t = '';
for (const a of arr) {
//logger.debug(`adding ${a.alternatives[0].transcript}`);
t += ` ${a.alternatives[0].transcript}`;
}
t += ` ${evt.alternatives[0].transcript}`;
evt.alternatives[0].transcript = t.trim();
//logger.debug(`compiled transcript: ${evt.alternatives[0].transcript}`);
};
class TaskGather extends Task {
@@ -29,6 +26,15 @@ class TaskGather extends Task {
super(logger, opts);
this.preconditions = TaskPreconditions.Endpoint;
const {
setChannelVarsForStt,
normalizeTranscription,
removeSpeechListeners
} = require('../utils/transcription-utils')(logger);
this.setChannelVarsForStt = setChannelVarsForStt;
this.normalizeTranscription = normalizeTranscription;
this.removeSpeechListeners = removeSpeechListeners;
[
'finishOnKey', 'hints', 'input', 'numDigits', 'minDigits', 'maxDigits',
'interDigitTimeout', 'partialResultHook', 'bargein', 'dtmfBargein',
@@ -47,47 +53,23 @@ class TaskGather extends Task {
const recognizer = this.data.recognizer;
this.vendor = recognizer.vendor;
this.language = recognizer.language;
this.hints = recognizer.hints || [];
this.hintsBoost = recognizer.hintsBoost;
this.profanityFilter = recognizer.profanityFilter;
this.punctuation = !!recognizer.punctuation;
this.enhancedModel = !!recognizer.enhancedModel;
this.model = recognizer.model || 'command_and_search';
this.words = !!recognizer.words;
this.singleUtterance = recognizer.singleUtterance || true;
this.diarization = !!recognizer.diarization;
this.diarizationMinSpeakers = recognizer.diarizationMinSpeakers || 0;
this.diarizationMaxSpeakers = recognizer.diarizationMaxSpeakers || 0;
this.interactionType = recognizer.interactionType || 'unspecified';
this.naicsCode = recognizer.naicsCode || 0;
this.altLanguages = recognizer.altLanguages || [];
if (recognizer.vendor === 'nuance') {
const {clientId, secret} = recognizer.nuanceOptions;
if (clientId && secret) {
this.sttCredentials = {client_id: clientId, secret};
}
}
/* continuous ASR (i.e. compile transcripts until a special timeout or dtmf key) */
this.asrTimeout = typeof recognizer.asrTimeout === 'number' ? recognizer.asrTimeout * 1000 : 0;
if (this.asrTimeout > 0) this.asrDtmfTerminationDigit = recognizer.asrDtmfTerminationDigit;
this.isContinuousAsr = this.asrTimeout > 0;
/* vad: if provided, we dont connect to recognizer until voice activity is detected */
const {enable, voiceMs = 0, mode = -1} = recognizer.vad || {};
this.vad = {enable, voiceMs, mode};
/* aws options */
this.vocabularyName = recognizer.vocabularyName;
this.vocabularyFilterName = recognizer.vocabularyFilterName;
this.filterMethod = recognizer.filterMethod;
/* microsoft options */
this.outputFormat = recognizer.outputFormat || 'simple';
this.profanityOption = recognizer.profanityOption || 'raw';
this.requestSnr = recognizer.requestSnr || false;
this.initialSpeechTimeoutMs = recognizer.initialSpeechTimeoutMs || 0;
this.azureServiceEndpoint = recognizer.azureServiceEndpoint;
this.azureSttEndpointId = recognizer.azureSttEndpointId;
}
else {
this.hints = [];
this.altLanguages = [];
this.data.recognizer.hints = this.data.recognizer.hints || [];
this.data.recognizer.altLanguages = this.data.recognizer.altLanguages || [];
}
else this.data.recognizer = {hints: [], altLanguages: []};
this.digitBuffer = '';
this._earlyMedia = this.data.earlyMedia === true;
@@ -134,21 +116,22 @@ class TaskGather extends Task {
this.logger.debug('Gather:exec');
await super.exec(cs);
const {updateSpeechCredentialLastUsed} = require('../utils/db-utils')(this.logger, cs.srf);
const {getNuanceAccessToken} = cs.srf.locals.dbHelpers;
if (cs.hasGlobalSttHints) {
const {hints, hintsBoost} = cs.globalSttHints;
this.hints = this.hints.concat(hints);
if (!this.hintsBoost && hintsBoost) this.hintsBoost = hintsBoost;
this.logger.debug({hints: this.hints, hintsBoost: this.hintsBoost},
this.data.recognizer.hints = this.data.recognizer.hints.concat(hints);
if (!this.data.recognizer.hintsBoost && hintsBoost) this.data.recognizer.hintsBoost = hintsBoost;
this.logger.debug({hints: this.data.recognizer.hints, hintsBoost: this.data.recognizer.hintsBoost},
'Gather:exec - applying global sttHints');
}
if (cs.hasAltLanguages) {
this.altLanguages = this.altLanguages.concat(cs.altLanguages);
this.data.recognizer.altLanguages = this.data.recognizer.altLanguages.concat(cs.altLanguages);
this.logger.debug({altLanguages: this.altLanguages},
'Gather:exec - applying altLanguages');
}
if (cs.hasGlobalSttPunctuation) {
this.punctuation = cs.globalSttPunctuation;
if (cs.hasGlobalSttPunctuation && !this.data.recognizer.punctuation) {
this.data.recognizer.punctuation = cs.globalSttPunctuation;
}
if (!this.isContinuousAsr && cs.isContinuousAsr) {
this.isContinuousAsr = true;
@@ -162,7 +145,8 @@ class TaskGather extends Task {
this.ep = ep;
if ('default' === this.vendor || !this.vendor) this.vendor = cs.speechRecognizerVendor;
if ('default' === this.language || !this.language) this.language = cs.speechRecognizerLanguage;
this.sttCredentials = cs.getSpeechCredentials(this.vendor, 'stt');
if (this.needsStt && !this.sttCredentials) this.sttCredentials = cs.getSpeechCredentials(this.vendor, 'stt');
if (this.needsStt && !this.sttCredentials) {
const {writeAlerts, AlertType} = cs.srf.locals;
this.logger.info(`TaskGather:exec - ERROR stt using ${this.vendor} requested but creds not supplied`);
@@ -175,16 +159,27 @@ class TaskGather extends Task {
throw new Error(`no speech-to-text service credentials for ${this.vendor} have been configured`);
}
this.logger.info({sttCredentials: this.sttCredentials}, 'Gather:exec - sttCredentials');
if (this.vendor === 'nuance' && this.sttCredentials.client_id) {
/* get nuance access token */
const {client_id, secret} = this.sttCredentials;
const {access_token, servedFromCache} = await getNuanceAccessToken(client_id, secret, 'asr tts');
this.logger.debug({client_id}, `Gather:exec - got nuance access token ${servedFromCache ? 'from cache' : ''}`);
this.sttCredentials = {...this.sttCredentials, access_token};
}
const startListening = (cs, ep) => {
this._startTimer();
if (this.isContinuousAsr && 0 === this.timeout) this._startAsrTimer();
if (this.input.includes('speech') && !this.listenDuringPrompt) {
this.logger.debug('Gather:exec - calling _initSpeech');
this._initSpeech(cs, ep)
.then(() => {
this._startTranscribing(ep);
return updateSpeechCredentialLastUsed(this.sttCredentials.speech_credential_sid);
})
.catch(() => {});
.catch((err) => {
this.logger.error({err}, 'error in initSpeech');
});
}
};
@@ -198,7 +193,15 @@ class TaskGather extends Task {
span.end();
if (err) this.logger.error({err}, 'Gather:exec Error playing tts');
this.logger.debug('Gather: nested say task completed');
if (!this.killed) startListening(cs, ep);
if (!this.killed) {
startListening(cs, ep);
if (this.input.includes('speech') && this.vendor === 'nuance' && this.listenDuringPrompt) {
this.logger.debug('Gather:exec - starting transcription timers after say completes');
ep.startTranscriptionTimers((err) => {
if (err) this.logger.error({err}, 'Gather:exec - error starting transcription timers');
});
}
}
});
}
else if (this.playTask) {
@@ -210,7 +213,15 @@ class TaskGather extends Task {
span.end();
if (err) this.logger.error({err}, 'Gather:exec Error playing url');
this.logger.debug('Gather: nested play task completed');
if (!this.killed) startListening(cs, ep);
if (!this.killed) {
startListening(cs, ep);
if (this.input.includes('speech') && this.vendor === 'nuance' && this.listenDuringPrompt) {
this.logger.debug('Gather:exec - starting transcription timers after play completes');
ep.startTranscriptionTimers((err) => {
if (err) this.logger.error({err}, 'Gather:exec - error starting transcription timers');
});
}
}
});
}
else startListening(cs, ep);
@@ -230,14 +241,7 @@ class TaskGather extends Task {
} catch (err) {
this.logger.error(err, 'TaskGather:exec error');
}
ep.removeCustomEventListener(GoogleTranscriptionEvents.Transcription);
ep.removeCustomEventListener(GoogleTranscriptionEvents.EndOfUtterance);
ep.removeCustomEventListener(GoogleTranscriptionEvents.VadDetected);
ep.removeCustomEventListener(AwsTranscriptionEvents.Transcription);
ep.removeCustomEventListener(AwsTranscriptionEvents.VadDetected);
ep.removeCustomEventListener(AzureTranscriptionEvents.Transcription);
ep.removeCustomEventListener(AzureTranscriptionEvents.NoSpeechDetected);
ep.removeCustomEventListener(AzureTranscriptionEvents.VadDetected);
this.removeSpeechListeners(ep);
}
kill(cs) {
@@ -292,106 +296,52 @@ class TaskGather extends Task {
}
async _initSpeech(cs, ep) {
const opts = {};
const opts = this.setChannelVarsForStt(this, this.sttCredentials, this.data.recognizer);
this.logger.debug(opts, 'TaskGather:_initSpeech - channel vars');
switch (this.vendor) {
case 'google':
this.bugname = 'google_transcribe';
ep.addCustomEventListener(GoogleTranscriptionEvents.Transcription, this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(GoogleTranscriptionEvents.EndOfUtterance, this._onEndOfUtterance.bind(this, cs, ep));
ep.addCustomEventListener(GoogleTranscriptionEvents.VadDetected, this._onVadDetected.bind(this, cs, ep));
break;
if (this.vad?.enable) {
opts.START_RECOGNIZING_ON_VAD = 1;
if (this.vad.voiceMs) opts.RECOGNIZER_VAD_VOICE_MS = this.vad.voiceMs;
else opts.RECOGNIZER_VAD_VOICE_MS = 125;
if (this.vad.mode >= 0 && this.vad.mode <= 3) opts.RECOGNIZER_VAD_MODE = this.vad.mode;
}
case 'aws':
case 'polly':
this.bugname = 'aws_transcribe';
ep.addCustomEventListener(AwsTranscriptionEvents.Transcription, this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(AwsTranscriptionEvents.VadDetected, this._onVadDetected.bind(this, cs, ep));
break;
case 'microsoft':
this.bugname = 'azure_transcribe';
ep.addCustomEventListener(AzureTranscriptionEvents.Transcription, this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(AzureTranscriptionEvents.NoSpeechDetected,
this._onNoSpeechDetected.bind(this, cs, ep));
ep.addCustomEventListener(AzureTranscriptionEvents.VadDetected, this._onVadDetected.bind(this, cs, ep));
break;
case 'nuance':
this.bugname = 'nuance_transcribe';
ep.addCustomEventListener(NuanceTranscriptionEvents.Transcription,
this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(NuanceTranscriptionEvents.StartOfSpeech,
this._onStartOfSpeech.bind(this, cs, ep));
ep.addCustomEventListener(NuanceTranscriptionEvents.TranscriptionComplete,
this._onTranscriptionComplete.bind(this, cs, ep));
ep.addCustomEventListener(NuanceTranscriptionEvents.VadDetected,
this._onVadDetected.bind(this, cs, ep));
ep.addCustomEventListener(NuanceTranscriptionEvents.Error,
this._onNuanceError.bind(this, cs, ep));
if ('google' === this.vendor) {
this.bugname = 'google_transcribe';
if (this.sttCredentials) opts.GOOGLE_APPLICATION_CREDENTIALS = JSON.stringify(this.sttCredentials.credentials);
[
['enhancedModel', 'GOOGLE_SPEECH_USE_ENHANCED'],
['separateRecognitionPerChannel', 'GOOGLE_SPEECH_SEPARATE_RECOGNITION_PER_CHANNEL'],
['profanityFilter', 'GOOGLE_SPEECH_PROFANITY_FILTER'],
['punctuation', 'GOOGLE_SPEECH_ENABLE_AUTOMATIC_PUNCTUATION'],
['words', 'GOOGLE_SPEECH_ENABLE_WORD_TIME_OFFSETS'],
['singleUtterance', 'GOOGLE_SPEECH_SINGLE_UTTERANCE'],
['diarization', 'GOOGLE_SPEECH_PROFANITY_FILTER']
].forEach((arr) => {
if (this[arr[0]]) opts[arr[1]] = true;
else if (this[arr[0]] === false) opts[arr[1]] = false;
});
if (this.hints.length > 0) {
opts.GOOGLE_SPEECH_HINTS = this.hints.join(',');
if (typeof this.hintsBoost === 'number') {
opts.GOOGLE_SPEECH_HINTS_BOOST = this.hintsBoost;
/* stall timers until prompt finishes playing */
if ((this.sayTask || this.playTask) && this.listenDuringPrompt) {
opts.NUANCE_STALL_TIMERS = 1;
}
}
if (this.altLanguages.length > 0) opts.GOOGLE_SPEECH_ALTERNATIVE_LANGUAGE_CODES = this.altLanguages.join(',');
else opts.GOOGLE_SPEECH_ALTERNATIVE_LANGUAGE_CODES = '';
if ('unspecified' !== this.interactionType) {
opts.GOOGLE_SPEECH_METADATA_INTERACTION_TYPE = this.interactionType;
}
opts.GOOGLE_SPEECH_MODEL = this.model;
if (this.diarization && this.diarizationMinSpeakers > 0) {
opts.GOOGLE_SPEECH_SPEAKER_DIARIZATION_MIN_SPEAKER_COUNT = this.diarizationMinSpeakers;
}
if (this.diarization && this.diarizationMaxSpeakers > 0) {
opts.GOOGLE_SPEECH_SPEAKER_DIARIZATION_MAX_SPEAKER_COUNT = this.diarizationMaxSpeakers;
}
if (this.naicsCode > 0) opts.GOOGLE_SPEECH_METADATA_INDUSTRY_NAICS_CODE = this.naicsCode;
ep.addCustomEventListener(GoogleTranscriptionEvents.Transcription, this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(GoogleTranscriptionEvents.EndOfUtterance, this._onEndOfUtterance.bind(this, cs, ep));
ep.addCustomEventListener(GoogleTranscriptionEvents.VadDetected, this._onVadDetected.bind(this, cs, ep));
}
else if (['aws', 'polly'].includes(this.vendor)) {
this.bugname = 'aws_transcribe';
if (this.vocabularyName) opts.AWS_VOCABULARY_NAME = this.vocabularyName;
if (this.vocabularyFilterName) {
opts.AWS_VOCABULARY_NAME = this.vocabularyFilterName;
opts.AWS_VOCABULARY_FILTER_METHOD = this.filterMethod || 'mask';
}
if (this.sttCredentials) {
Object.assign(opts, {
AWS_ACCESS_KEY_ID: this.sttCredentials.accessKeyId,
AWS_SECRET_ACCESS_KEY: this.sttCredentials.secretAccessKey,
AWS_REGION: this.sttCredentials.region
});
}
ep.addCustomEventListener(AwsTranscriptionEvents.Transcription, this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(AwsTranscriptionEvents.VadDetected, this._onVadDetected.bind(this, cs, ep));
}
else if ('microsoft' === this.vendor) {
this.bugname = 'azure_transcribe';
if (this.sttCredentials) {
const {api_key, region, use_custom_stt, custom_stt_endpoint} = this.sttCredentials;
Object.assign(opts, {
'AZURE_SUBSCRIPTION_KEY': api_key,
'AZURE_REGION': region
});
if (this.azureSttEndpointId) {
Object.assign(opts, {'AZURE_SERVICE_ENDPOINT_ID': this.azureSttEndpointId});
}
else if (use_custom_stt && custom_stt_endpoint) {
Object.assign(opts, {'AZURE_SERVICE_ENDPOINT_ID': custom_stt_endpoint});
}
}
if (this.hints && this.hints.length > 0) {
opts.AZURE_SPEECH_HINTS = this.hints.map((h) => h.trim()).join(',');
}
if (this.altLanguages && this.altLanguages.length > 0) {
opts.AZURE_SPEECH_ALTERNATIVE_LANGUAGE_CODES = this.altLanguages.join(',');
}
else {
opts.AZURE_SPEECH_ALTERNATIVE_LANGUAGE_CODES = '';
}
if (this.requestSnr) opts.AZURE_REQUEST_SNR = 1;
if (this.profanityOption && this.profanityOption !== 'raw') opts.AZURE_PROFANITY_OPTION = this.profanityOption;
if (this.azureServiceEndpoint) opts.AZURE_SERVICE_ENDPOINT = this.azureServiceEndpoint;
if (this.initialSpeechTimeoutMs > 0) opts.AZURE_INITIAL_SPEECH_TIMEOUT_MS = this.initialSpeechTimeoutMs;
else if (this.timeout === 0) opts.AZURE_INITIAL_SPEECH_TIMEOUT_MS = 120000; // lengthy
opts.AZURE_USE_OUTPUT_FORMAT_DETAILED = 1;
ep.addCustomEventListener(AzureTranscriptionEvents.Transcription, this._onTranscription.bind(this, cs, ep));
ep.addCustomEventListener(AzureTranscriptionEvents.NoSpeechDetected, this._onNoSpeechDetected.bind(this, cs, ep));
ep.addCustomEventListener(AzureTranscriptionEvents.VadDetected, this._onVadDetected.bind(this, cs, ep));
break;
default:
throw new Error(`Invalid vendor ${this.vendor}`);
}
await ep.set(opts)
.catch((err) => this.logger.info(err, 'Error setting channel variables'));
}
@@ -489,40 +439,12 @@ class TaskGather extends Task {
_onTranscription(cs, ep, evt, fsEvent) {
// make sure this is not a transcript from answering machine detection
this.logger.debug({evt}, 'Gather:_onTranscription');
const bugname = fsEvent.getHeader('media-bugname');
const finished = fsEvent.getHeader('transcription-session-finished');
if (bugname && this.bugname !== bugname) return;
if ('aws' === this.vendor && Array.isArray(evt) && evt.length > 0) evt = evt[0];
if ('microsoft' === this.vendor) {
const final = evt.RecognitionStatus === 'Success';
if (final) {
// don't sort based on confidence: https://github.com/Azure-Samples/cognitive-services-speech-sdk/issues/1463
//const nbest = evt.NBest.sort((a, b) => b.Confidence - a.Confidence);
const nbest = evt.NBest;
const language_code = evt.PrimaryLanguage?.Language || this.language;
evt = {
is_final: true,
language_code,
alternatives: [
{
confidence: nbest[0].Confidence,
transcript: nbest[0].Display
}
]
};
}
else {
evt = {
is_final: false,
alternatives: [
{
transcript: evt.Text
}
]
};
}
}
evt = this.normalizeTranscription(evt, this.vendor, 1, this.language);
/* count words for bargein feature */
const words = evt.alternatives[0].transcript.split(' ').length;
@@ -607,6 +529,24 @@ class TaskGather extends Task {
}
}
_onStartOfSpeech(cs, ep) {
this.logger.debug('TaskGather:_onStartOfSpeech');
}
_onTranscriptionComplete(cs, ep) {
this.logger.debug('TaskGather:_onTranscriptionComplete');
}
_onNuanceError(cs, ep, evt) {
const {code, error, details} = evt;
if (code === 404 && error === 'No speech') {
this.logger.debug({code, error, details}, 'TaskGather:_onNuanceError');
return this._resolve('timeout');
}
this.logger.info({code, error, details}, 'TaskGather:_onNuanceError');
if (code === 413 && error === 'Too much speech') {
return this._resolve('timeout');
}
}
_onVadDetected(cs, ep) {
if (this.bargein && this.minBargeinWordCount === 0) {
this.logger.debug('TaskGather:_onVadDetected');