Files
jambonz-api-server/lib/utils/speech-utils.js
2023-12-18 12:37:10 -05:00

423 lines
13 KiB
JavaScript

const sttGoogle = require('@google-cloud/speech').v1p1beta1;
const { TranscribeClient, ListVocabulariesCommand } = require('@aws-sdk/client-transcribe');
const { Deepgram } = require('@deepgram/sdk');
const sdk = require('microsoft-cognitiveservices-speech-sdk');
const { SpeechClient } = require('@soniox/soniox-node');
const bent = require('bent');
const fs = require('fs');
const { AssemblyAI } = require('assemblyai');
const {decrypt, obscureKey} = require('./encrypt-decrypt');
const testSonioxStt = async(logger, credentials) => {
const api_key = credentials;
const soniox = new SpeechClient(api_key);
return new Promise(async(resolve, reject) => {
try {
const result = await soniox.transcribeFileShort('data/test_audio.wav');
if (result.words.length > 0) resolve(result);
else reject(new Error('no transcript returned'));
} catch (error) {
logger.info({error}, 'failed to get soniox transcript');
reject(error);
}
});
};
const testNuanceTts = async(logger, getTtsVoices, credentials) => {
const voices = await getTtsVoices({vendor: 'nuance', credentials});
return voices;
};
const testNuanceStt = async(logger, credentials) => {
//TODO
return true;
};
const testGoogleTts = async(logger, getTtsVoices, credentials) => {
const voices = await getTtsVoices({vendor: 'google', credentials});
return voices;
};
const testGoogleStt = async(logger, credentials) => {
const client = new sttGoogle.SpeechClient({credentials});
const config = {
sampleRateHertz: 8000,
languageCode: 'en-US',
model: 'default',
};
const audio = {
content: fs.readFileSync(`${__dirname}/../../data/test_audio.wav`).toString('base64'),
};
const request = {
config: config,
audio: audio,
};
// Detects speech in the audio file
const [response] = await client.recognize(request);
if (!Array.isArray(response.results) || 0 === response.results.length) {
throw new Error('failed to transcribe speech');
}
};
const testDeepgramStt = async(logger, credentials) => {
const {api_key} = credentials;
const deepgram = new Deepgram(api_key);
const mimetype = 'audio/wav';
const source = {
buffer: fs.readFileSync(`${__dirname}/../../data/test_audio.wav`),
mimetype: mimetype
};
return new Promise((resolve, reject) => {
// Send the audio to Deepgram and get the response
deepgram.transcription
.preRecorded(source, {punctuate: true})
.then((response) => {
//logger.debug({response}, 'got transcript');
if (response?.results?.channels[0]?.alternatives?.length > 0) resolve(response);
else reject(new Error('no transcript returned'));
return;
})
.catch((err) => {
logger.info({err}, 'failed to get deepgram transcript');
reject(err);
});
});
};
const testMicrosoftStt = async(logger, credentials) => {
const {api_key, region} = credentials;
const speechConfig = sdk.SpeechConfig.fromSubscription(api_key, region);
const audioConfig = sdk.AudioConfig.fromWavFileInput(fs.readFileSync(`${__dirname}/../../data/test_audio.wav`));
speechConfig.speechRecognitionLanguage = 'en-US';
if (process.env.JAMBONES_HTTP_PROXY_IP && process.env.JAMBONES_HTTP_PROXY_PORT) {
logger.debug(
`testMicrosoftStt: using proxy ${process.env.JAMBONES_HTTP_PROXY_IP}:${process.env.JAMBONES_HTTP_PROXY_PORT}`);
speechConfig.setProxy(process.env.JAMBONES_HTTP_PROXY_IP, process.env.JAMBONES_HTTP_PROXY_PORT);
}
const speechRecognizer = new sdk.SpeechRecognizer(speechConfig, audioConfig);
return new Promise((resolve, reject) => {
speechRecognizer.recognizeOnceAsync((result) => {
switch (result.reason) {
case sdk.ResultReason.RecognizedSpeech:
resolve();
break;
case sdk.ResultReason.NoMatch:
reject('Speech could not be recognized.');
break;
case sdk.ResultReason.Canceled:
const cancellation = sdk.CancellationDetails.fromResult(result);
logger.info(`CANCELED: Reason=${cancellation.reason}`);
if (cancellation.reason == sdk.CancellationReason.Error) {
logger.info(`CANCELED: ErrorCode=${cancellation.ErrorCode}`);
logger.info(`CANCELED: ErrorDetails=${cancellation.errorDetails}`);
}
reject(cancellation.reason);
break;
}
speechRecognizer.close();
});
});
};
const testAwsTts = async(logger, getTtsVoices, credentials) => {
try {
const voices = await getTtsVoices({vendor: 'aws', credentials});
return voices;
} catch (err) {
logger.info({err}, 'testMicrosoftTts - failed to list voices for region ${region}');
throw err;
}
};
const testAwsStt = async(logger, credentials) => {
try {
const {region, accessKeyId, secretAccessKey} = credentials;
const client = new TranscribeClient({
region,
credentials: {
accessKeyId,
secretAccessKey
}
});
const command = new ListVocabulariesCommand({});
const response = await client.send(command);
return response;
} catch (err) {
logger.info({err}, 'testMicrosoftTts - failed to list voices for region ${region}');
throw err;
}
};
const testMicrosoftTts = async(logger, synthAudio, credentials) => {
try {
await synthAudio({increment: () => {}, histogram: () => {}},
{
vendor: 'microsoft',
credentials,
language: 'en-US',
voice: 'en-US-JennyMultilingualNeural',
text: 'Hi there and welcome to jambones!'
}
);
} catch (err) {
logger.info({err}, 'testMicrosoftTts returned error');
throw err;
}
};
const testWellSaidTts = async(logger, credentials) => {
const {api_key} = credentials;
try {
const post = bent('https://api.wellsaidlabs.com', 'POST', 'buffer', {
'X-Api-Key': api_key,
'Accept': 'audio/mpeg',
'Content-Type': 'application/json'
});
const mp3 = await post('/v1/tts/stream', {
text: 'Hello, world',
speaker_id: '3'
});
return mp3;
} catch (err) {
logger.info({err}, 'testWellSaidTts returned error');
throw err;
}
};
const testElevenlabs = async(logger, credentials) => {
const {api_key, model_id} = credentials;
try {
const post = bent('https://api.elevenlabs.io', 'POST', 'buffer', {
'xi-api-key': api_key,
'Accept': 'audio/mpeg',
'Content-Type': 'application/json'
});
const mp3 = await post('/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM', {
text: 'Hello',
model_id,
voice_settings: {
stability: 0.5,
similarity_boost: 0.5
}
});
return mp3;
} catch (err) {
logger.info({err}, 'synthEvenlabs returned error');
throw err;
}
};
const testWhisper = async(logger, synthAudio, credentials) => {
try {
await synthAudio({increment: () => {}, histogram: () => {}},
{
vendor: 'whisper',
credentials,
language: 'en-US',
voice: 'alloy',
text: 'Hi there and welcome to jambones!'
}
);
} catch (err) {
logger.info({err}, 'synthEvenlabs returned error');
throw err;
}
};
const testIbmTts = async(logger, getTtsVoices, credentials) => {
const {tts_api_key, tts_region} = credentials;
const voices = await getTtsVoices({vendor: 'ibm', credentials: {tts_api_key, tts_region}});
return voices;
};
const testIbmStt = async(logger, credentials) => {
const {stt_api_key, stt_region} = credentials;
const SpeechToTextV1 = require('ibm-watson/speech-to-text/v1');
const { IamAuthenticator } = require('ibm-watson/auth');
const speechToText = new SpeechToTextV1({
authenticator: new IamAuthenticator({
apikey: stt_api_key
}),
serviceUrl: `https://api.${stt_region}.speech-to-text.watson.cloud.ibm.com`
});
return new Promise((resolve, reject) => {
speechToText.listModels()
.then((speechModels) => {
logger.debug({speechModels}, 'got IBM speech models');
return resolve();
})
.catch((err) => {
logger.info({err}, 'failed to get speech models');
reject(err);
});
});
};
const testWellSaidStt = async(logger, credentials) => {
//TODO
return true;
};
const testAssemblyStt = async(logger, credentials) => {
const {api_key} = credentials;
const assemblyai = new AssemblyAI({
apiKey: api_key
});
const audioUrl = `${__dirname}/../../data/test_audio.wav`;
return new Promise((resolve, reject) => {
assemblyai.transcripts
.create({ audio_url: audioUrl })
.then((transcript) => {
logger.debug({transcript}, 'got transcription from AssemblyAi');
if (transcript.status === 'error') {
return reject({message: transcript.error});
}
return resolve(transcript.text);
})
.catch((err) => {
logger.info({err}, 'failed to get assemblyAI transcription');
reject(err);
});
});
};
const getSpeechCredential = (credential, logger) => {
const {vendor} = credential;
logger.info(
`Speech vendor: ${credential.vendor} ${credential.label ? `, label: ${credential.label}` : ''} selected`);
if ('google' === vendor) {
try {
const cred = JSON.parse(credential.service_key.replace(/\n/g, '\\n'));
return {
...credential,
credentials: cred
};
} catch (err) {
logger.info({err}, `malformed google service_key provisioned for account ${credential.speech_credential_sid}`);
}
}
else if (['aws', 'polly'].includes(vendor)) {
return {
...credential,
accessKeyId: credential.access_key_id,
secretAccessKey: credential.secret_access_key,
region: credential.aws_region || 'us-east-1'
};
}
return credential;
};
function decryptCredential(obj, credential, logger, isObscureKey = true) {
if ('google' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
const key_header = '-----BEGIN PRIVATE KEY-----\n';
const obscured = {
...o,
private_key: `${key_header}${isObscureKey ?
obscureKey(o.private_key.slice(key_header.length, o.private_key.length)) :
o.private_key.slice(key_header.length, o.private_key.length)}`
};
obj.service_key = JSON.stringify(obscured);
}
else if ('aws' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.access_key_id = o.access_key_id;
obj.secret_access_key = isObscureKey ? obscureKey(o.secret_access_key) : o.secret_access_key;
obj.aws_region = o.aws_region;
logger.info({obj, o}, 'retrieving aws speech credential');
}
else if ('microsoft' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
obj.region = o.region;
obj.use_custom_tts = o.use_custom_tts;
obj.custom_tts_endpoint = o.custom_tts_endpoint;
obj.custom_tts_endpoint_url = o.custom_tts_endpoint_url;
obj.use_custom_stt = o.use_custom_stt;
obj.custom_stt_endpoint = o.custom_stt_endpoint;
obj.custom_stt_endpoint_url = o.custom_stt_endpoint_url;
logger.info({obj, o}, 'retrieving azure speech credential');
}
else if ('wellsaid' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
}
else if ('nuance' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.client_id = o.client_id;
obj.secret = o.secret ? (isObscureKey ? obscureKey(o.secret) : o.secret) : null;
}
else if ('deepgram' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
}
else if ('ibm' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.tts_api_key = isObscureKey ? obscureKey(o.tts_api_key) : o.tts_api_key;
obj.tts_region = o.tts_region;
obj.stt_api_key = isObscureKey ? obscureKey(o.stt_api_key) : o.stt_api_key;
obj.stt_region = o.stt_region;
obj.instance_id = o.instance_id;
} else if ('nvidia' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.riva_server_uri = o.riva_server_uri;
} else if ('cobalt' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.cobalt_server_uri = o.cobalt_server_uri;
} else if ('soniox' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
} else if ('elevenlabs' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
obj.model_id = o.model_id;
obj.options = o.options;
} else if (obj.vendor.startsWith('custom:')) {
const o = JSON.parse(decrypt(credential));
obj.auth_token = isObscureKey ? obscureKey(o.auth_token) : o.auth_token;
obj.custom_stt_url = o.custom_stt_url;
obj.custom_tts_url = o.custom_tts_url;
} else if ('assemblyai' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
} else if ('whisper' === obj.vendor) {
const o = JSON.parse(decrypt(credential));
obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
obj.model_id = o.model_id;
}
}
module.exports = {
testGoogleTts,
testGoogleStt,
testAwsTts,
testWellSaidTts,
testAwsStt,
testMicrosoftTts,
testMicrosoftStt,
testWellSaidStt,
testNuanceTts,
testNuanceStt,
testDeepgramStt,
testIbmTts,
testIbmStt,
testSonioxStt,
testElevenlabs,
testAssemblyStt,
getSpeechCredential,
decryptCredential,
testWhisper
};