bugfix: AWS polly

fix: custom tts support multiple audio types

lib/purge-tts-cache.js

@@ -11,7 +11,8 @@ const debug = require('debug')('jambonz:realtimedb-helpers');
  * @param {string} opts.text - text or ssml to synthesize
  * @returns {object} result - {error, purgedCount}
  */
-async function purgeTtsCache(client, logger, {all, vendor, language, voice, deploymentId, engine, text} = {all: true}) {
+async function purgeTtsCache(client, logger, {all, account_sid, vendor,
+  language, voice, deploymentId, engine, text} = {all: true}) {
   logger = logger || noopLogger;
 
   let purgedCount = 0, error;
@@ -21,8 +22,13 @@ async function purgeTtsCache(client, logger, {all, vendor, language, voice, depl
       const keys = await client.keysAsync('tts:*');
       purgedCount = await client.delAsync(keys);
 
-    } else {
+    } else if (account_sid && !vendor && !language && !voice && !engine && !text) {
+      const keys = await client.keysAsync(`tts:${account_sid}:*`);
+      purgedCount = await client.delAsync(keys);
+    }
+    else {
       const key = makeSynthKey({
+        account_sid,
         vendor,
         language: language || '',
         voice: voice || deploymentId,

lib/synth-audio.js

@@ -32,7 +32,7 @@ const {
 const {SynthesizeSpeechRequest} = require('../stubs/riva/proto/riva_tts_pb');
 const {AudioEncoding} = require('../stubs/riva/proto/riva_audio_pb');
 const debug = require('debug')('jambonz:realtimedb-helpers');
-const EXPIRES = 3600 * 24; // cache tts for 24 hours
+const EXPIRES = (process.env.JAMBONES_TTS_CACHE_DURATION_MINS || 24 * 60) * 60; // cache tts for 24 hours
 const TMP_FOLDER = '/tmp';
 
 /**
@@ -52,7 +52,7 @@ const TMP_FOLDER = '/tmp';
  * @returns object containing filepath to an mp3 file in the /tmp folder containing
  * the synthesized audio, and a variable indicating whether it was served from cache
  */
-async function synthAudio(client, logger, stats, {
+async function synthAudio(client, logger, stats, { account_sid,
   vendor, language, voice, gender, text, engine, salt, model, credentials, deploymentId, disableTtsCache
 }) {
   let audioBuffer;
@@ -95,8 +95,8 @@ async function synthAudio(client, logger, stats, {
   } else  if (vendor.startsWith('custom')) {
     assert.ok(credentials.custom_tts_url, `synthAudio requires custom_tts_url in credentials when ${vendor} is used`);
   }
-
   const key = makeSynthKey({
+    account_sid,
     vendor,
     language: language || '',
     voice: voice || deploymentId,
@@ -155,7 +155,8 @@ async function synthAudio(client, logger, stats, {
         audioBuffer = await synthWellSaid(logger, {credentials, stats, language, voice, text, filePath});
         break;
       case vendor.startsWith('custom') ? vendor : 'cant_match_value':
-        audioBuffer = await synthCustomVendor(logger, {credentials, stats, language, voice, text});
+        ({ audioBuffer, filePath } = await synthCustomVendor(logger,
+          {credentials, stats, language, voice, text, filePath}));
         break;
       default:
         assert(`synthAudio: unsupported speech vendor ${vendor}`);
@@ -183,7 +184,7 @@ async function synthAudio(client, logger, stats, {
 
 const synthPolly = async(logger, {credentials, stats, language, voice, engine, text}) => {
   try {
-    const polly = new PollyClient(credentials);
+    const polly = new PollyClient({credentials});
     const opts = {
       Engine: engine,
       OutputFormat: 'mp3',
@@ -440,30 +441,56 @@ const synthNvidia = async(client, logger, {credentials, stats, language,  voice,
 };
 
 
-// CustomVendor accept only mp3
-const synthCustomVendor = async(logger, {credentials, stats, language, voice, text}) => {
+const synthCustomVendor = async(logger, {credentials, stats, language, voice, text, filePath}) => {
   const {vendor, auth_token, custom_tts_url} = credentials;
 
   try {
-    const post = bent('POST', 'buffer', {
+    const post = bent('POST', {
       'Authorization': `Bearer ${auth_token}`,
-      'Accept': 'audio/mpeg',
       'Content-Type': 'application/json'
     });
 
-    const mp3 = await post(custom_tts_url, {
+    const response = await post(custom_tts_url, {
       language,
-      format: 'audio/mpeg',
       voice,
       type: text.startsWith('<speak>') ? 'ssml' : 'text',
       text
     });
 
-    return mp3;
+    const regex = /\.[^\.]*$/g;
+    const mime = response.headers['content-type'];
+    const buffer = await response.arrayBuffer();
+    return {
+      audioBuffer: buffer,
+      filePath: filePath.replace(regex, getFileExtFromMime(mime))
+    };
   } catch (err) {
     logger.info({err}, `Vendor ${vendor} returned error`);
     throw err;
   }
 };
 
+const getFileExtFromMime = (mime) => {
+  switch (mime) {
+    case 'audio/wav':
+    case 'audio/x-wav':
+      return '.wav';
+    case /audio\/l16.*rate=8000/.test(mime) ? mime : 'cant match value':
+      return '.r8';
+    case /audio\/l16.*rate=16000/.test(mime) ? mime : 'cant match value':
+      return '.r16';
+    case /audio\/l16.*rate=24000/.test(mime) ? mime : 'cant match value':
+      return '.r24';
+    case /audio\/l16.*rate=32000/.test(mime) ? mime : 'cant match value':
+      return '.r32';
+    case /audio\/l16.*rate=48000/.test(mime) ? mime : 'cant match value':
+      return '.r48';
+    case 'audio/mpeg':
+    case 'audio/mp3':
+      return '.mp3';
+    default:
+      return '.wav';
+  }
+};
+
 module.exports = synthAudio;

lib/utils.js

@@ -16,10 +16,10 @@ const debug = require('debug')('jambonz:realtimedb-helpers');
  */
 //const nuanceClientMap = new Map();
 
-function makeSynthKey({vendor, language, voice, engine = '', text}) {
+function makeSynthKey({account_sid = '', vendor, language, voice, engine = '', text}) {
   const hash = crypto.createHash('sha1');
   hash.update(`${language}:${vendor}:${voice}:${engine}:${text}`);
-  return `tts:${hash.digest('hex')}`;
+  return `tts${account_sid ? (':' + account_sid) : ''}:${hash.digest('hex')}`;
 }
 
 const noopLogger = {

package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "@jambonz/speech-utils",
-  "version": "0.0.1",
+  "version": "0.0.4",
   "lockfileVersion": 2,
   "requires": true,
   "packages": {
     "": {
       "name": "@jambonz/speech-utils",
-      "version": "0.0.1",
+      "version": "0.0.4",
       "license": "MIT",
       "dependencies": {
         "@aws-sdk/client-polly": "^3.276.0",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jambonz/speech-utils",
-  "version": "0.0.1",
+  "version": "0.0.4",
   "description": "TTS-related speech utilities for jambonz",
   "main": "index.js",
   "author": "Dave Horton",

test/synth.js

@@ -348,7 +348,6 @@ test('Custom Vendor speech synth tests', async(t) => {
     let obj = await getJSON(`http://127.0.0.1:3100/lastRequest/somethingnew`);
     t.ok(obj.headers.Authorization == 'Bearer some_jwt_token', 'Custom Vendor Authentication Header is correct');
     t.ok(obj.body.language == 'en-US', 'Custom Vendor Language is correct');
-    t.ok(obj.body.format == 'audio/mpeg', 'Custom Vendor format is correct');
     t.ok(obj.body.voice == 'English-US.Female-1', 'Custom Vendor voice is correct');
     t.ok(obj.body.type == 'text', 'Custom Vendor type is correct');
     t.ok(obj.body.text == 'This is a test.  This is only a test', 'Custom Vendor text is correct');
@@ -427,5 +426,35 @@ test('TTS Cache tests', async(t) => {
     t.end(err);
   }
 
+  try {
+    // clear cache
+    await purgeTtsCache();
+    // save some random tts keys to cache
+    const minRecords = 8;
+    const account_sid = "12412512_cabc_5aff"
+    const account_sid2 = "22412512_cabc_5aff"
+    for (const i in Array(minRecords).fill(0)) {
+      await client.setAsync(makeSynthKey({account_sid, vendor: i, language: i, voice: i, engine: i, text: i}), i);
+    }
+    for (const i in Array(minRecords).fill(0)) {
+      await client.setAsync(makeSynthKey({account_sid: account_sid2, vendor: i, language: i, voice: i, engine: i, text: i}), i);
+    }
+    const {purgedCount} = await purgeTtsCache({account_sid});
+    t.equal(purgedCount, minRecords, `successfully purged at least ${minRecords} tts records from cache for account_sid:${account_sid}`);
+
+    let cached = (await client.keysAsync('tts:*')).length;
+    t.equal(cached, minRecords, `successfully purged all tts records from cache for account_sid:${account_sid}`);
+
+    const {purgedCount: purgedCount2} = await purgeTtsCache({account_sid: account_sid2});
+    t.equal(purgedCount2, minRecords, `successfully purged at least ${minRecords} tts records from cache for account_sid:${account_sid2}`);
+
+    cached = (await client.keysAsync('tts:*')).length;
+    t.equal(cached, 0, `successfully purged all tts records from cache`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+
   client.quit();
 });