playht should return list of voice match voice engine configured at speech credentials

* support mod_playht_tts

* wip

* wip

* wip

* wip

* wip

* update speech utils version

db/jambones-sql.sql

@@ -162,7 +162,7 @@ regex VARCHAR(32) NOT NULL COMMENT 'regex-based pattern match against dialed num
 description VARCHAR(1024),
 priority INTEGER NOT NULL COMMENT 'lower priority routes are attempted first',
 PRIMARY KEY (lcr_route_sid)
-) COMMENT='An ordered list of  digit patterns in an LCR table.  The pat';
+) COMMENT='An ordered list of  digit patterns in an LCR table.  The patterns are tested in sequence until one matches';
 
 CREATE TABLE lcr
 (
@@ -173,7 +173,7 @@ default_carrier_set_entry_sid CHAR(36) COMMENT 'default carrier/route to use whe
 service_provider_sid CHAR(36),
 account_sid CHAR(36),
 PRIMARY KEY (lcr_sid)
-) COMMENT='An LCR (least cost routing) table that is used by a service ';
+) COMMENT='An LCR (least cost routing) table that is used by a service provider or account to make decisions about routing outbound calls when multiple carriers are available.';
 
 CREATE TABLE password_settings
 (
@@ -458,6 +458,7 @@ inbound BOOLEAN NOT NULL COMMENT 'if true, whitelist this IP to allow inbound ca
 outbound BOOLEAN NOT NULL COMMENT 'if true, include in least-cost routing when placing calls to the PSTN',
 voip_carrier_sid CHAR(36) NOT NULL,
 is_active BOOLEAN NOT NULL DEFAULT 1,
+send_options_ping BOOLEAN NOT NULL DEFAULT 0,
 pad_crypto BOOLEAN NOT NULL DEFAULT 0,
 protocol ENUM('udp','tcp','tls', 'tls/srtp') DEFAULT 'udp' COMMENT 'Outbound call protocol',
 PRIMARY KEY (sip_gateway_sid)
@@ -495,7 +496,7 @@ messaging_hook_sid CHAR(36) COMMENT 'webhook to call for inbound SMS/MMS ',
 app_json TEXT,
 speech_synthesis_vendor VARCHAR(64) NOT NULL DEFAULT 'google',
 speech_synthesis_language VARCHAR(12) NOT NULL DEFAULT 'en-US',
-speech_synthesis_voice VARCHAR(64),
+speech_synthesis_voice VARCHAR(256),
 speech_synthesis_label VARCHAR(64),
 speech_recognizer_vendor VARCHAR(64) NOT NULL DEFAULT 'google',
 speech_recognizer_language VARCHAR(64) NOT NULL DEFAULT 'en-US',
@@ -503,7 +504,7 @@ speech_recognizer_label VARCHAR(64),
 use_for_fallback_speech BOOLEAN DEFAULT false,
 fallback_speech_synthesis_vendor VARCHAR(64),
 fallback_speech_synthesis_language VARCHAR(12),
-fallback_speech_synthesis_voice VARCHAR(64),
+fallback_speech_synthesis_voice VARCHAR(256),
 fallback_speech_synthesis_label VARCHAR(64),
 fallback_speech_recognizer_vendor VARCHAR(64),
 fallback_speech_recognizer_language VARCHAR(64),

db/jambones.sqs

@@ -551,7 +551,7 @@
         </location>
         <size>
             <width>293.00</width>
-            <height>540.00</height>
+            <height>560.00</height>
         </size>
         <zorder>6</zorder>
         <SQLField>
@@ -2332,7 +2332,7 @@
         </location>
         <size>
             <width>281.00</width>
-            <height>240.00</height>
+            <height>260.00</height>
         </size>
         <zorder>7</zorder>
         <SQLField>
@@ -2399,10 +2399,18 @@
             <notNull><![CDATA[1]]></notNull>
             <uid><![CDATA[27D4A5BD-8093-4ADD-B5B5-D546844206F9]]></uid>
         </SQLField>
+        <SQLField>
+            <name><![CDATA[send_options_ping]]></name>
+            <type><![CDATA[BOOLEAN]]></type>
+            <defaultValue><![CDATA[0]]></defaultValue>
+            <notNull><![CDATA[1]]></notNull>
+            <uid><![CDATA[E04C19A2-12BF-443F-AB61-96990224A18D]]></uid>
+        </SQLField>
         <SQLField>
             <name><![CDATA[pad_crypto]]></name>
             <type><![CDATA[BOOLEAN]]></type>
             <defaultValue><![CDATA[0]]></defaultValue>
+            <forcedUnique><![CDATA[0]]></forcedUnique>
             <notNull><![CDATA[1]]></notNull>
             <uid><![CDATA[C5C0043B-100A-4476-BF01-BE0777AE27C0]]></uid>
         </SQLField>
@@ -2560,7 +2568,7 @@
         </SQLField>
         <SQLField>
             <name><![CDATA[speech_synthesis_voice]]></name>
-            <type><![CDATA[VARCHAR(64)]]></type>
+            <type><![CDATA[VARCHAR(256)]]></type>
             <notNull><![CDATA[0]]></notNull>
             <uid><![CDATA[929D66F0-64B9-4D7C-AB4B-24F131E1178F]]></uid>
         </SQLField>
@@ -2610,7 +2618,7 @@
         </SQLField>
         <SQLField>
             <name><![CDATA[fallback_speech_synthesis_voice]]></name>
-            <type><![CDATA[VARCHAR(64)]]></type>
+            <type><![CDATA[VARCHAR(256)]]></type>
             <notNull><![CDATA[0]]></notNull>
             <uid><![CDATA[6A0E92C9-32B9-4179-A893-3DADF5DD7728]]></uid>
         </SQLField>
@@ -3097,17 +3105,17 @@
         <overviewPanelHidden><![CDATA[0]]></overviewPanelHidden>
         <pageBoundariesVisible><![CDATA[0]]></pageBoundariesVisible>
         <PageGridVisible><![CDATA[0]]></PageGridVisible>
-        <RightSidebarWidth><![CDATA[1924.000000]]></RightSidebarWidth>
+        <RightSidebarWidth><![CDATA[1235.000000]]></RightSidebarWidth>
         <sidebarIndex><![CDATA[2]]></sidebarIndex>
         <snapToGrid><![CDATA[0]]></snapToGrid>
-        <SourceSidebarWidth><![CDATA[0.000000]]></SourceSidebarWidth>
+        <SourceSidebarWidth><![CDATA[312.000000]]></SourceSidebarWidth>
         <SQLEditorFileFormatVersion><![CDATA[4]]></SQLEditorFileFormatVersion>
         <uid><![CDATA[58C99A00-06C9-478C-A667-C63842E088F3]]></uid>
-        <windowHeight><![CDATA[985.000000]]></windowHeight>
-        <windowLocationX><![CDATA[-1307.000000]]></windowLocationX>
-        <windowLocationY><![CDATA[1008.000000]]></windowLocationY>
-        <windowScrollOrigin><![CDATA[{1.5, 786}]]></windowScrollOrigin>
-        <windowWidth><![CDATA[2201.000000]]></windowWidth>
+        <windowHeight><![CDATA[870.000000]]></windowHeight>
+        <windowLocationX><![CDATA[-1164.000000]]></windowLocationX>
+        <windowLocationY><![CDATA[1131.000000]]></windowLocationY>
+        <windowScrollOrigin><![CDATA[{0.5, 0}]]></windowScrollOrigin>
+        <windowWidth><![CDATA[1512.000000]]></windowWidth>
     </SQLDocumentInfo>
     <AllowsIndexRenamingOnInsert><![CDATA[1]]></AllowsIndexRenamingOnInsert>
     <defaultLabelExpanded><![CDATA[1]]></defaultLabelExpanded>

db/upgrade-jambonz-db.js

@@ -190,7 +190,12 @@ const sql = {
     'ALTER TABLE google_custom_voices ADD FOREIGN KEY speech_credential_sid_idxfk (speech_credential_sid) REFERENCES speech_credentials (speech_credential_sid) ON DELETE CASCADE',
     'ALTER TABLE clients ADD COLUMN allow_direct_queue_calling BOOLEAN NOT NULL DEFAULT 1',
     'ALTER TABLE clients ADD COLUMN allow_direct_user_calling BOOLEAN NOT NULL DEFAULT 1',
-    'ALTER TABLE clients ADD COLUMN allow_direct_app_calling BOOLEAN NOT NULL DEFAULT 1'
+    'ALTER TABLE clients ADD COLUMN allow_direct_app_calling BOOLEAN NOT NULL DEFAULT 1',
+  ],
+  9000: [
+    'ALTER TABLE sip_gateways ADD COLUMN send_options_ping BOOLEAN NOT NULL DEFAULT 0',
+    'ALTER TABLE applications MODIFY COLUMN speech_synthesis_voice VARCHAR(256)',
+    'ALTER TABLE applications MODIFY COLUMN fallback_speech_synthesis_voice VARCHAR(256)',
   ]
 };
 
@@ -223,6 +228,7 @@ const doIt = async() => {
         if (val < 8003) upgrades.push(...sql['8003']);
         if (val < 8004) upgrades.push(...sql['8004']);
         if (val < 8005) upgrades.push(...sql['8005']);
+        if (val < 9000) upgrades.push(...sql['9000']);
 
         // perform all upgrades
         logger.info({upgrades}, 'applying schema upgrades..');

lib/models/voip-carrier.js

@@ -61,6 +61,10 @@ VoipCarrier.fields = [
     name: 'requires_register',
     type: 'number'
   },
+  {
+    name: 'register_use_tls',
+    type: 'number'
+  },
   {
     name: 'register_username',
     type: 'string'

lib/routes/api/speech-credentials.js

@@ -6,7 +6,8 @@ const sysError = require('../error');
 const {decrypt, encrypt} = require('../../utils/encrypt-decrypt');
 const {parseAccountSid, parseServiceProviderSid, parseSpeechCredentialSid} = require('./utils');
 const {decryptCredential, testWhisper, testDeepgramTTS,
-  getLanguagesAndVoicesForVendor} = require('../../utils/speech-utils');
+  getLanguagesAndVoicesForVendor,
+  testPlayHT} = require('../../utils/speech-utils');
 const {DbErrorUnprocessableRequest, DbErrorForbidden, DbErrorBadRequest} = require('../../utils/errors');
 const {
   testGoogleTts,
@@ -135,6 +136,8 @@ const encryptCredential = (obj) => {
     auth_token = '',
     cobalt_server_uri,
     model_id,
+    user_id,
+    voice_engine,
     options
   } = obj;
 
@@ -219,6 +222,13 @@ const encryptCredential = (obj) => {
       const elevenlabsData = JSON.stringify({api_key, model_id, options});
       return encrypt(elevenlabsData);
 
+    case 'playht':
+      assert(api_key, 'invalid playht speech credential: api_key is required');
+      assert(user_id, 'invalid playht speech credential: user_id is required');
+      assert(voice_engine, 'invalid voice_engine speech credential: voice_engine is required');
+      const playhtData = JSON.stringify({api_key, user_id, voice_engine, options});
+      return encrypt(playhtData);
+
     case 'assemblyai':
       assert(api_key, 'invalid assemblyai speech credential: api_key is required');
       const assemblyaiData = JSON.stringify({api_key});
@@ -418,6 +428,7 @@ router.put('/:sid', async(req, res) => {
           custom_tts_url,
           cobalt_server_uri,
           model_id,
+          voice_engine,
           options,
           deepgram_stt_uri,
           deepgram_stt_use_tls,
@@ -443,6 +454,7 @@ router.put('/:sid', async(req, res) => {
           custom_tts_url,
           cobalt_server_uri,
           model_id,
+          voice_engine,
           options,
           deepgram_stt_uri,
           deepgram_stt_use_tls,
@@ -724,6 +736,17 @@ router.get('/:sid/test', async(req, res) => {
           SpeechCredential.ttsTestResult(sid, false);
         }
       }
+    } else if (cred.vendor === 'playht') {
+      if (cred.use_for_tts) {
+        try {
+          await testPlayHT(logger, synthAudio, credential);
+          results.tts.status = 'ok';
+          SpeechCredential.ttsTestResult(sid, true);
+        } catch (err) {
+          results.tts = {status: 'fail', reason: err.message};
+          SpeechCredential.ttsTestResult(sid, false);
+        }
+      }
     } else if (cred.vendor === 'assemblyai') {
       const {api_key} = credential;
       if (cred.use_for_stt) {

lib/swagger/swagger.yaml

@@ -1108,6 +1108,9 @@ paths:
                 requires_register:
                   type: boolean
                   description: wehther this provider requires us to send a REGISTER to them in order to receive calls
+                register_use_tls:
+                  type: boolean
+                  description: wehther this provider requires us to send a REGISTER use TLS protocol
                 register_username:
                   type: string
                   description: sip username to authenticate with, if registration is required

lib/utils/speech-data/tts-model-playht.js

@@ -0,0 +1,6 @@
+module.exports = [
+  { name: 'PlayHT2.0-turbo', value: 'PlayHT2.0-turbo' },
+  { name: 'PlayHT2.0', value: 'PlayHT2.0' },
+  { name: 'PlayHT1.0', value: 'PlayHT1.0' },
+];
+

lib/utils/speech-utils.js

@@ -21,6 +21,7 @@ const TtsWhisperLanguagesVoices = require('./speech-data/tts-whisper');
 const TtsModelDeepgram = require('./speech-data/tts-model-deepgram');
 const TtsModelElevenLabs = require('./speech-data/tts-model-elevenlabs');
 const TtsModelWhisper = require('./speech-data/tts-model-whisper');
+const TtsModelPlayHT = require('./speech-data/tts-model-playht');
 
 const SttGoogleLanguagesVoices = require('./speech-data/stt-google');
 const SttAwsLanguagesVoices = require('./speech-data/stt-aws');
@@ -240,6 +241,27 @@ const testElevenlabs = async(logger, credentials) => {
   }
 };
 
+const testPlayHT = async(logger, synthAudio, credentials) => {
+  try {
+    await synthAudio(
+      {
+        increment: () => {},
+        histogram: () => {}
+      },
+      {
+        vendor: 'playht',
+        credentials,
+        language: 'en-US',
+        voice: 's3://voice-cloning-zero-shot/d9ff78ba-d016-47f6-b0ef-dd630f59414e/female-cs/manifest.json',
+        text: 'Hi there and welcome to jambones!'
+      }
+    );
+  } catch (err) {
+    logger.info({err}, 'synth Playht returned error');
+    throw err;
+  }
+};
+
 const testWhisper = async(logger, synthAudio, credentials) => {
   try {
     await synthAudio({increment: () => {}, histogram: () => {}},
@@ -428,6 +450,12 @@ function decryptCredential(obj, credential, logger, isObscureKey = true) {
     obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
     obj.model_id = o.model_id;
     obj.options = o.options;
+  } else if ('playht' === obj.vendor) {
+    const o = JSON.parse(decrypt(credential));
+    obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
+    obj.user_id = o.user_id;
+    obj.voice_engine = o.voice_engine;
+    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;
@@ -488,6 +516,8 @@ async function getLanguagesAndVoicesForVendor(logger, vendor, credential, getTts
       return await getLanguagesVoicesForSoniox(credential, getTtsVoices, logger);
     case 'elevenlabs':
       return await getLanguagesVoicesForElevenlabs(credential, getTtsVoices, logger);
+    case 'playht':
+      return await getLanguagesVoicesForPlayHT(credential, getTtsVoices, logger);
     case 'assemblyai':
       return await getLanguagesVoicesForAssemblyAI(credential, getTtsVoices, logger);
     case 'whisper':
@@ -645,6 +675,52 @@ async function getLanguagesVoicesForElevenlabs(credential) {
   }
 }
 
+const concat = (a) => {
+  return a ? ` ${a},` : '';
+};
+
+async function getLanguagesVoicesForPlayHT(credential) {
+  if (credential) {
+    const get = bent('https://api.play.ht', 'GET', 'json', {
+      'AUTHORIZATION' : credential.api_key,
+      'X-USER-ID': credential.user_id,
+      'Accept': 'application/json'
+    });
+
+    const voices = await get('/api/v2/voices');
+
+    const buildVoice = (d) => {
+      let name = `${d.name} -${concat(d.accent)}${concat(d.age)}${concat(d.gender)}
+${concat(d.loudness)}${concat(d.style)}${concat(d.tempo)}${concat(d.texture)}` ;
+      name = name.endsWith(',') ? name.slice(0, -1) : name;
+      return {
+        value: `${d.id}`,
+        name
+      };
+    };
+
+    const ttsVoices = voices.reduce((acc, voice) => {
+      if (!credential.voice_engine.includes(voice.voice_engine)) {
+        return acc;
+      }
+      const languageCode = voice.language_code;
+      const existingLanguage = acc.find((lang) => lang.value === languageCode);
+      if (existingLanguage) {
+        existingLanguage.voices.push(buildVoice(voice));
+      } else {
+        acc.push({
+          value: voice.language_code,
+          name: voice.language,
+          voices: [buildVoice(voice)]
+        });
+      }
+      return acc;
+    }, []);
+    return tranform(ttsVoices, undefined, TtsModelPlayHT);
+  }
+  return tranform(undefined, undefined, TtsModelPlayHT);
+}
+
 async function getLanguagesVoicesForAssemblyAI(credential) {
   return tranform(undefined, SttAssemblyaiLanguagesVoices);
 }
@@ -796,6 +872,7 @@ module.exports = {
   testIbmStt,
   testSonioxStt,
   testElevenlabs,
+  testPlayHT,
   testAssemblyStt,
   testDeepgramTTS,
   getSpeechCredential,

package.json

@@ -1,6 +1,6 @@
 {
   "name": "jambonz-api-server",
-  "version": "0.8.5",
+  "version": "0.9.0",
   "description": "",
   "main": "app.js",
   "scripts": {
@@ -19,54 +19,54 @@
     "url": "https://github.com/jambonz/jambonz-api-server.git"
   },
   "dependencies": {
-    "@aws-sdk/client-s3": "^3.363.0",
-    "@aws-sdk/client-transcribe": "^3.363.0",
-    "@azure/storage-blob": "^12.15.0",
+    "@aws-sdk/client-s3": "^3.550.0",
+    "@aws-sdk/client-transcribe": "^3.549.0",
+    "@azure/storage-blob": "^12.17.0",
     "@deepgram/sdk": "^1.21.0",
-    "@google-cloud/speech": "^5.2.0",
-    "@google-cloud/storage": "^6.12.0",
+    "@google-cloud/speech": "^6.5.0",
+    "@google-cloud/storage": "^7.9.0",
     "@jambonz/db-helpers": "^0.9.3",
     "@jambonz/lamejs": "^1.2.2",
     "@jambonz/mw-registrar": "^0.2.7",
-    "@jambonz/realtimedb-helpers": "^0.8.7",
-    "@jambonz/speech-utils": "^0.0.41",
+    "@jambonz/realtimedb-helpers": "^0.8.8",
+    "@jambonz/speech-utils": "^0.0.50",
     "@jambonz/time-series": "^0.2.8",
-    "@jambonz/verb-specifications": "^0.0.45",
-    "@soniox/soniox-node": "^1.1.1",
-    "argon2": "^0.30.3",
-    "assemblyai": "^3.0.1",
+    "@jambonz/verb-specifications": "^0.0.69",
+    "@soniox/soniox-node": "^1.2.2",
+    "argon2": "^0.40.1",
+    "assemblyai": "^4.3.4",
     "bent": "^7.3.12",
     "cors": "^2.8.5",
     "debug": "^4.3.4",
-    "express": "^4.18.1",
-    "express-rate-limit": "^6.4.0",
-    "form-data": "^2.5.1",
-    "helmet": "^5.1.0",
-    "ibm-watson": "^7.1.2",
-    "jsonwebtoken": "^9.0.0",
-    "mailgun.js": "^9.1.2",
-    "microsoft-cognitiveservices-speech-sdk": "1.31.0",
-    "mysql2": "^2.3.3",
-    "nocache": "3.0.4",
-    "passport": "^0.6.0",
+    "express": "^4.19.2",
+    "express-rate-limit": "^7.2.0",
+    "form-data": "^4.0.0",
+    "helmet": "^7.1.0",
+    "ibm-watson": "^9.0.1",
+    "jsonwebtoken": "^9.0.2",
+    "mailgun.js": "^10.2.1",
+    "microsoft-cognitiveservices-speech-sdk": "1.36.0",
+    "mysql2": "^3.9.3",
+    "nocache": "4.0.0",
+    "passport": "^0.7.0",
     "passport-http-bearer": "^1.0.1",
-    "pino": "^5.17.0",
-    "short-uuid": "^4.1.0",
+    "pino": "^8.20.0",
+    "short-uuid": "^4.2.2",
     "stream-buffers": "^3.0.2",
-    "stripe": "^8.222.0",
-    "swagger-ui-express": "^4.4.0",
-    "uuid": "^8.3.2",
+    "stripe": "^14.24.0",
+    "swagger-ui-express": "^5.0.0",
+    "uuid": "^9.0.1",
     "wav": "^1.0.2",
-    "ws": "^8.12.1",
+    "ws": "^8.16.0",
     "yamljs": "^0.3.0"
   },
   "devDependencies": {
     "eslint": "^8.39.0",
     "eslint-plugin-promise": "^6.1.1",
-    "husky": "7.0.4",
+    "husky": "9.0.11",
     "nyc": "^15.1.0",
     "request": "^2.88.2",
     "request-promise-native": "^1.0.9",
-    "tape": "^5.5.3"
+    "tape": "^5.7.5"
   }
 }

test/docker-compose-testbed.yaml

@@ -10,7 +10,7 @@ networks:
 
 services:  
   mysql:
-    platform: linux/x86_64
+    # platform: linux/x86_64
     image: mysql:5.7
     ports:
       - "3360:3306"
@@ -36,7 +36,7 @@ services:
         ipv4_address: 172.58.0.3
     
   influxdb:
-    platform: linux/x86_64
+    # platform: linux/x86_64
     image: influxdb:1.8
     ports:
       - "8086:8086"

test/speech-credentials.js

@@ -536,7 +536,7 @@ test('speech credentials tests', async(t) => {
         model_id: 'eleven_multilingual_v2'
       }
     });
-    t.ok(result.statusCode === 201, 'successfully added speech credential for Cobalt');
+    t.ok(result.statusCode === 201, 'successfully added speech credential for elevenlabs');
     const elevenlabs_sid = result.body.sid;
 
     /* delete the credential */
@@ -544,7 +544,31 @@ test('speech credentials tests', async(t) => {
       auth: authUser,
       resolveWithFullResponse: true,
     });
-    t.ok(result.statusCode === 204, 'successfully deleted speech credential for Cobalt');
+    t.ok(result.statusCode === 204, 'successfully deleted speech credential for elevenlabs');
+
+    /* add a credential for playht */
+    result = await request.post(`/Accounts/${account_sid}/SpeechCredentials`, {
+      resolveWithFullResponse: true,
+      auth: authUser,
+      json: true,
+      body: {
+        vendor: 'playht',
+        use_for_stt: false,
+        use_for_tts: true,
+        api_key: 'asdasdasdasddsadasda',
+        user_id: 'user_id',
+        voice_engine: 'PlayHT2.0-turbo'
+      }
+    });
+    t.ok(result.statusCode === 201, 'successfully added speech credential for playht');
+    const playht_sid = result.body.sid;
+
+    /* delete the credential */
+    result = await request.delete(`/Accounts/${account_sid}/SpeechCredentials/${playht_sid}`, {
+      auth: authUser,
+      resolveWithFullResponse: true,
+    });
+    t.ok(result.statusCode === 204, 'successfully deleted speech credential for playht');
 
 
     /* add a credential for custom voices google */