0.2.11

Feat/893 azure ssml

lib/add-file-to-cache.js

@@ -3,8 +3,29 @@ const {noopLogger, makeSynthKey} = require('./utils');
 const {JAMBONES_TTS_CACHE_DURATION_MINS} = require('./config');
 const EXPIRES = JAMBONES_TTS_CACHE_DURATION_MINS;
 
+function getExtensionAndSampleRate(path) {
+  const match = path.match(/\.([^.]*)$/);
+  if (!match) {
+    //default should be wav file.
+    return ['wav', 8000];
+  }
+
+  const extension = match[1];
+  const sampleRateMap = {
+    r8: 8000,
+    r16: 16000,
+    r24: 24000,
+    r44: 44100,
+    r48: 48000,
+    r96: 96000,
+  };
+
+  const sampleRate = sampleRateMap[extension] || 8000;
+  return [extension, sampleRate];
+}
+
 async function addFileToCache(client, logger, path,
-  {account_sid, vendor, language, voice, deploymentId, engine, text}) {
+  {account_sid, vendor, language, voice, deploymentId, engine, model, text, instructions}) {
   let key;
   logger = logger || noopLogger;
 
@@ -15,10 +36,19 @@ async function addFileToCache(client, logger, path,
       language: language || '',
       voice: voice || deploymentId,
       engine,
+      model,
       text,
+      instructions
     });
+    const [extension, sampleRate] = getExtensionAndSampleRate(path);
     const audioBuffer = await fs.readFile(path);
-    await client.setex(key, EXPIRES, audioBuffer.toString('base64'));
+    await client.setex(key, EXPIRES, JSON.stringify(
+      {
+        audioContent: audioBuffer.toString('base64'),
+        extension,
+        sampleRate
+      }
+    ));
   } catch (err) {
     logger.error(err, 'addFileToCache: Error');
     return;

lib/config.js

@@ -2,6 +2,7 @@ const JAMBONES_TTS_TRIM_SILENCE = process.env.JAMBONES_TTS_TRIM_SILENCE;
 const JAMBONES_DISABLE_TTS_STREAMING = process.env.JAMBONES_DISABLE_TTS_STREAMING;
 const JAMBONES_DISABLE_AZURE_TTS_STREAMING = process.env.JAMBONES_DISABLE_AZURE_TTS_STREAMING;
 const JAMBONES_EAGERLY_PRE_CACHE_AUDIO = process.env.JAMBONES_EAGERLY_PRE_CACHE_AUDIO;
+const JAMBONES_AZURE_ENABLE_SSML = process.env.JAMBONES_AZURE_ENABLE_SSML;
 
 const JAMBONES_HTTP_PROXY_IP = process.env.JAMBONES_HTTP_PROXY_IP;
 const JAMBONES_HTTP_PROXY_PORT = process.env.JAMBONES_HTTP_PROXY_PORT;
@@ -21,5 +22,6 @@ module.exports = {
   JAMBONES_TTS_CACHE_DURATION_MINS,
   JAMBONES_EAGERLY_PRE_CACHE_AUDIO,
   TMP_FOLDER,
-  HTTP_TIMEOUT
+  HTTP_TIMEOUT,
+  JAMBONES_AZURE_ENABLE_SSML
 };

lib/purge-tts-cache.js

@@ -12,7 +12,7 @@ const debug = require('debug')('jambonz:realtimedb-helpers');
  * @returns {object} result - {error, purgedCount}
  */
 async function purgeTtsCache(client, logger, {all, account_sid, vendor,
-  language, voice, deploymentId, engine, text} = {all: true}) {
+  language, voice, deploymentId, engine, model, text, instructions} = {all: true}) {
   logger = logger || noopLogger;
 
   let purgedCount = 0, error;
@@ -33,7 +33,9 @@ async function purgeTtsCache(client, logger, {all, account_sid, vendor,
         language: language || '',
         voice: voice || deploymentId,
         engine,
+        model,
         text,
+        instructions
       });
       purgedCount = await client.del(key);
       if (purgedCount === 0) error = 'Specified item not found';

lib/synth-audio.js

@@ -46,7 +46,8 @@ const {
   JAMBONES_HTTP_PROXY_IP,
   JAMBONES_HTTP_PROXY_PORT,
   JAMBONES_TTS_CACHE_DURATION_MINS,
-  JAMBONES_EAGERLY_PRE_CACHE_AUDIO,
+  JAMBONES_TTS_TRIM_SILENCE,
+  JAMBONES_AZURE_ENABLE_SSML
 } = require('./config');
 const EXPIRES = JAMBONES_TTS_CACHE_DURATION_MINS;
 const OpenAI = require('openai');
@@ -89,9 +90,9 @@ const trimTrailingSilence = (buffer) => {
  */
 async function synthAudio(client, createHash, retrieveHash, logger, stats, { account_sid,
   vendor, language, voice, gender, text, engine, salt, model, credentials, deploymentId,
-  disableTtsCache, renderForCaching = false, disableTtsStreaming, options
+  disableTtsCache, renderForCaching = false, disableTtsStreaming, options, instructions
 }) {
-  let audioBuffer;
+  let audioData;
   let servedFromCache = false;
   let rtt;
   logger = logger || noopLogger;
@@ -169,57 +170,27 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
     language: language || '',
     voice: voice || deploymentId,
     engine,
+    // model or model_id is used to identify the tts cache.
+    model: model || credentials.model_id,
     text,
-    renderForCaching
+    instructions
   });
-  let filePath;
-  // used only for custom vendor
-  let fileExtension;
-  filePath = makeFilePath({vendor, voice, key, salt, renderForCaching});
+
   debug(`synth key is ${key}`);
   let cached;
   if (!disableTtsCache) {
     cached = await client.get(key);
-    /**
-    *  If we are using tts streaming and also precaching audio, audio could have been cached by streaming (r8)
-    * or here in speech-utils due to precaching (mp3), so we need to check for both keys.
-    */
-    if (!cached && JAMBONES_EAGERLY_PRE_CACHE_AUDIO) {
-      const preCachekey = makeSynthKey({
-        account_sid,
-        vendor,
-        language: language || '',
-        voice: voice || deploymentId,
-        engine,
-        text,
-        renderForCaching: true
-      });
-      cached = await client.get(preCachekey);
-      if (cached) {
-        // Precache audio is available update filpath with precache file extension.
-        filePath = makeFilePath({vendor, voice, key, salt, renderForCaching: true});
-      }
-    }
   }
   if (cached) {
     // found in cache - extend the expiry and use it
     debug('result WAS found in cache');
     servedFromCache = true;
     stats.increment('tts.cache.requests', ['found:yes']);
-    if (vendor.startsWith('custom')) {
-      // custom vendors support multiple mime types such as: mp3, wav, r8, r16 ...etc,
-      // mime type/file extension is available when http response has header Content-type.
-      // In cache, file extension is store together with audiBuffer in a json.
-      // Normal cache audio will be base64 string
-      const payload = JSON.parse(cached);
-      filePath = filePath.replace(/\.[^\.]*$/g, payload.fileExtension);
-      audioBuffer = Buffer.from(payload.audioBuffer, 'base64');
-    } else {
-      audioBuffer = Buffer.from(cached, 'base64');
-    }
+    audioData = JSON.parse(cached);
+    // convert base64 audio to buffer
+    audioData.audioContent = Buffer.from(audioData.audioContent, 'base64');
     client.expire(key, EXPIRES).catch((err) => logger.info(err, 'Error setting expires'));
-  }
-  if (!cached) {
+  } else {
     // not found in cache - go get it from speech vendor and add to cache
     debug('result was NOT found in cache');
     stats.increment('tts.cache.requests', ['found:no']);
@@ -227,94 +198,93 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
     const startAt = process.hrtime();
     switch (vendor) {
       case 'google':
-        audioBuffer = await synthGoogle(logger, {credentials, stats, language, voice, gender, text});
+        audioData = await synthGoogle(logger, {credentials, stats, language, voice, gender, text});
         break;
       case 'aws':
       case 'polly':
         vendorLabel = 'aws';
-        audioBuffer = await synthPolly(createHash, retrieveHash, logger,
+        audioData = await synthPolly(createHash, retrieveHash, logger,
           {credentials, stats, language, voice, text, engine});
         break;
       case 'azure':
       case 'microsoft':
         vendorLabel = 'microsoft';
-        audioBuffer = await synthMicrosoft(logger, {credentials, stats, language, voice, text, deploymentId,
-          filePath, renderForCaching, disableTtsStreaming});
+        audioData = await synthMicrosoft(logger, {credentials, stats, language, voice, text, deploymentId,
+          renderForCaching, disableTtsStreaming});
         break;
       case 'nuance':
         model = model || 'enhanced';
-        audioBuffer = await synthNuance(client, logger, {credentials, stats, voice, model, text});
+        audioData = await synthNuance(client, logger, {credentials, stats, voice, model, text});
         break;
       case 'nvidia':
-        audioBuffer = await synthNvidia(client, logger, {credentials, stats, language, voice, model, text});
+        audioData = await synthNvidia(client, logger, {credentials, stats, language, voice, model, text,
+          renderForCaching, disableTtsStreaming});
         break;
       case 'ibm':
-        audioBuffer = await synthIbm(logger, {credentials, stats, voice, text});
+        audioData = await synthIbm(logger, {credentials, stats, voice, text});
         break;
       case 'wellsaid':
-        audioBuffer = await synthWellSaid(logger, {credentials, stats, language, voice, text, filePath});
+        audioData = await synthWellSaid(logger, {credentials, stats, language, voice, text});
         break;
       case 'elevenlabs':
-        audioBuffer = await synthElevenlabs(logger, {
-          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming, filePath
-        });
+        audioData = await synthElevenlabs(logger, {
+          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming});
         break;
       case 'playht':
-        audioBuffer = await synthPlayHT(client, logger, {
-          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming, filePath
-        });
+        audioData = await synthPlayHT(client, logger, {
+          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming});
         break;
       case 'cartesia':
-        audioBuffer = await synthCartesia(logger, {
-          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming, filePath
-        });
+        audioData = await synthCartesia(logger, {
+          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming});
         break;
       case 'rimelabs':
-        audioBuffer = await synthRimelabs(logger, {
-          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming, filePath
-        });
+        audioData = await synthRimelabs(logger, {
+          credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming});
         break;
       case 'whisper':
-        audioBuffer = await synthWhisper(logger, {
-          credentials, stats, voice, text, renderForCaching, disableTtsStreaming});
+        audioData = await synthWhisper(logger, {
+          credentials, stats, voice, text, instructions, renderForCaching, disableTtsStreaming});
         break;
       case 'verbio':
-        audioBuffer = await synthVerbio(client, logger, {
+        audioData = await synthVerbio(client, logger, {
           credentials, stats, voice, text, renderForCaching, disableTtsStreaming});
-        if (audioBuffer?.filePath) return audioBuffer;
+        if (audioData?.filePath) return audioData;
         break;
       case 'deepgram':
-        audioBuffer = await synthDeepgram(logger, {credentials, stats, model, text,
+        audioData = await synthDeepgram(logger, {credentials, stats, model, text,
           renderForCaching, disableTtsStreaming});
         break;
       case vendor.startsWith('custom') ? vendor : 'cant_match_value':
-        ({ audioBuffer, filePath, fileExtension } = await synthCustomVendor(logger,
-          {credentials, stats, language, voice, text, filePath}));
+        audioData = await synthCustomVendor(logger,
+          {credentials, stats, language, voice, text});
         break;
       default:
         assert(`synthAudio: unsupported speech vendor ${vendor}`);
     }
-    if ('filePath' in audioBuffer) return audioBuffer;
+    if ('filePath' in audioData) return audioData;
     const diff = process.hrtime(startAt);
     const time = diff[0] * 1e3 + diff[1] * 1e-6;
     rtt = time.toFixed(0);
     stats.histogram('tts.response_time', rtt, [`vendor:${vendorLabel}`]);
     debug(`tts rtt time for ${text.length} chars on ${vendorLabel}: ${rtt}`);
     logger.info(`tts rtt time for ${text.length} chars on ${vendorLabel}: ${rtt}`);
-
-    const base64Audio = audioBuffer.toString('base64');
-    const cacheContent = vendor.startsWith('custom') ?
-      JSON.stringify({
-        audioBuffer: base64Audio,
-        fileExtension
-      }) : base64Audio;
-
-    client.setex(key, EXPIRES, cacheContent)
+    // Save audio json to cache
+    client.setex(key, EXPIRES, JSON.stringify({
+      ...audioData,
+      audioContent: audioData.audioContent?.toString('base64')
+    }))
       .catch((err) => logger.error(err, `error calling setex on key ${key}`));
   }
 
   return new Promise((resolve, reject) => {
-    fs.writeFile(filePath, audioBuffer, (err) => {
+    const { audioContent, extension } = audioData;
+    const filePath = makeFilePath({
+      key,
+      salt,
+      extension
+    });
+    fs.writeFile(filePath, audioContent, (err) => {
       if (err) return reject(err);
       resolve({filePath, servedFromCache, rtt});
     });
@@ -369,7 +339,13 @@ const synthPolly = async(createHash, retrieveHash, logger,
         .on('data', (chunk) => {
           chunks.push(chunk);
         })
-        .on('end', () => resolve(Buffer.concat(chunks)));
+        .on('end', () => resolve(
+          {
+            audioContent: Buffer.concat(chunks),
+            extension: 'mp3',
+            sampleRate: 8000
+          }
+        ));
     });
   } catch (err) {
     logger.info({err}, 'synthAudio: Error synthesizing speech using aws polly');
@@ -411,7 +387,11 @@ const synthGoogle = async(logger, {credentials, stats, language, voice, gender,
       };
 
       const wav = await post('/v1beta1/text:synthesize', payload);
-      return Buffer.from(wav.audioContent, 'base64');
+      return {
+        audioContent: Buffer.from(wav.audioContent, 'base64'),
+        extension: 'wav',
+        sampleRate: 24000
+      };
     } catch (err) {
       logger.info({err: await err.text()}, 'synthGoogle returned error');
       throw err;
@@ -432,7 +412,11 @@ const synthGoogle = async(logger, {credentials, stats, language, voice, gender,
     const responses = await client.synthesizeSpeech(opts);
     stats.increment('tts.count', ['vendor:google', 'accepted:yes']);
     client.close();
-    return responses[0].audioContent;
+    return {
+      audioContent: responses[0].audioContent,
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     console.error(err);
     logger.info({err, opts}, 'synthAudio: Error synthesizing speech using google');
@@ -463,7 +447,11 @@ const synthIbm = async(logger, {credentials, stats, voice, text}) => {
     for await (const chunk of r.result) {
       chunks.push(chunk);
     }
-    return Buffer.concat(chunks);
+    return {
+      audioContent: Buffer.concat(chunks),
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err, params}, 'synthAudio: Error synthesizing speech using ibm');
     stats.increment('tts.count', ['vendor:ibm', 'accepted:no']);
@@ -473,15 +461,12 @@ const synthIbm = async(logger, {credentials, stats, voice, text}) => {
 
 async function _synthOnPremMicrosoft(logger, {
   credentials,
-  stats,
   language,
   voice,
-  text,
-  filePath
+  text
 }) {
   const {use_custom_tts, custom_tts_endpoint_url, api_key} = credentials;
   let content = text;
-
   if (use_custom_tts && !content.startsWith('<speak')) {
     /**
      * Note: it seems that to use custom voice ssml is required with the voice attribute
@@ -494,20 +479,28 @@ async function _synthOnPremMicrosoft(logger, {
     /* microsoft enforces some properties and uses voice xml element so if the user did not supply do it for them */
     const words = content.slice(7, -8).trim().replace(/(\r\n|\n|\r)/gm, ' ');
     // eslint-disable-next-line max-len
-    content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><voice name="${voice}">${words}</voice></speak>`;
+    content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><lang xml:lang="${language}"><voice name="${voice}">${words}</voice></lang></speak>`;
     logger.info({content}, 'synthMicrosoft');
   }
+  else if (JAMBONES_AZURE_ENABLE_SSML) {
+    // eslint-disable-next-line max-len
+    content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><voice name="${voice}"><lang xml:lang="${language}">${text}</lang></voice></speak>`;
+  }
 
   try {
-    const trimSilence = filePath.endsWith('.r8');
+    const trimSilence = JAMBONES_TTS_TRIM_SILENCE;
     const post = bent('POST', 'buffer', {
       'X-Microsoft-OutputFormat': trimSilence ? 'raw-8khz-16bit-mono-pcm' : 'audio-16khz-32kbitrate-mono-mp3',
       'Content-Type': 'application/ssml+xml',
       'User-Agent': 'Jambonz',
       ...(api_key && {'Ocp-Apim-Subscription-Key': api_key})
     });
-    const mp3 = await post(custom_tts_endpoint_url, content);
-    return mp3;
+    const audioContent = await post(custom_tts_endpoint_url, content);
+    return {
+      audioContent,
+      extension: trimSilence ? 'r8' : 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, '_synthMicrosoftByHttp returned error');
     throw err;
@@ -520,7 +513,6 @@ const synthMicrosoft = async(logger, {
   language,
   voice,
   text,
-  filePath,
   renderForCaching,
   disableTtsStreaming
 }) => {
@@ -530,19 +522,23 @@ const synthMicrosoft = async(logger, {
     let content = text;
     if (use_custom_tts && !content.startsWith('<speak')) {
       /**
-       * Note: it seems that to use custom voice ssml is required with the voice attribute
-       * Otherwise sending plain text we get "Voice does not match"
-       */
+     * Note: it seems that to use custom voice ssml is required with the voice attribute
+     * Otherwise sending plain text we get "Voice does not match"
+     */
       content = `<speak>${text}</speak>`;
     }
 
     if (content.startsWith('<speak>')) {
-      /* microsoft enforces some properties and uses voice xml element so if the user did not supply do it for them */
+    /* microsoft enforces some properties and uses voice xml element so if the user did not supply do it for them */
       const words = content.slice(7, -8).trim().replace(/(\r\n|\n|\r)/gm, ' ');
       // eslint-disable-next-line max-len
-      content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><voice name="${voice}">${words}</voice></speak>`;
+      content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><lang xml:lang="${language}"><voice name="${voice}">${words}</voice></lang></speak>`;
       logger.info({content}, 'synthMicrosoft');
     }
+    else if (JAMBONES_AZURE_ENABLE_SSML) {
+      // eslint-disable-next-line max-len
+      content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><voice name="${voice}"><lang xml:lang="${language}">${text}</lang></voice></speak>`;
+    }
     if (!JAMBONES_DISABLE_TTS_STREAMING && !JAMBONES_DISABLE_AZURE_TTS_STREAMING &&
       !renderForCaching && !disableTtsStreaming) {
       let params = '';
@@ -563,17 +559,18 @@ const synthMicrosoft = async(logger, {
         rtt: 0
       };
     }
+    // Azure Onprem
     if (use_custom_tts && custom_tts_endpoint_url) {
       return await _synthOnPremMicrosoft(logger, {
         credentials,
         stats,
         language,
         voice,
-        text,
-        filePath
+        text
       });
     }
-    const trimSilence = filePath.endsWith('.r8');
+    // Azure hosted service
+    const trimSilence = JAMBONES_TTS_TRIM_SILENCE;
     const speechConfig = SpeechConfig.fromSubscription(apiKey, region);
     speechConfig.speechSynthesisLanguage = language;
     speechConfig.speechSynthesisVoiceName = voice;
@@ -608,7 +605,11 @@ const synthMicrosoft = async(logger, {
             case ResultReason.SynthesizingAudioCompleted:
               let buffer = Buffer.from(result.audioData);
               if (trimSilence) buffer = trimTrailingSilence(buffer);
-              resolve(buffer);
+              resolve({
+                audioContent: buffer,
+                extension: trimSilence ? 'r8' : 'mp3',
+                sampleRate: 8000
+              });
               synthesizer.close();
               stats.increment('tts.count', ['vendor:microsoft', 'accepted:yes']);
               break;
@@ -638,11 +639,15 @@ const synthWellSaid = async(logger, {credentials, stats, language, voice, gender
       'Accept': 'audio/mpeg',
       'Content-Type': 'application/json'
     });
-    const mp3 = await post('/v1/tts/stream', {
+    const audioContent = await post('/v1/tts/stream', {
       text,
       speaker_id: voice
     });
-    return mp3;
+    return {
+      audioContent,
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, 'testWellSaidTts returned error');
     throw err;
@@ -679,8 +684,8 @@ const synthNuance = async(client, logger, {credentials, stats, voice, model, tex
     t.setText(text);
     input.setText(t);
   }
-
-  pcm.setSampleRateHz(8000);
+  const sampleRate = 8000;
+  pcm.setSampleRateHz(sampleRate);
   f.setPcm(pcm);
   p.setAudioFormat(f);
   v.setName(voice);
@@ -704,20 +709,41 @@ const synthNuance = async(client, logger, {credentials, stats, voice, model, tex
         const details = status.getDetails();
         return reject({code, message, details});
       }
-      resolve(Buffer.from(response.getAudio()));
+      resolve({
+        audioContent: Buffer.from(response.getAudio()),
+        extension: 'r8',
+        sampleRate
+      });
     });
   });
 };
 
-const synthNvidia = async(client, logger, {credentials, stats, language,  voice, model, text}) => {
+const synthNvidia = async(client, logger, {
+  credentials, stats, language,  voice, model, text, renderForCaching, disableTtsStreaming
+}) => {
   const {riva_server_uri} = credentials;
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    let params = '';
+    params += `{riva_server_uri=${riva_server_uri}`;
+    params += `,voice=${voice}`;
+    params += `,language=${language}`;
+    params += ',write_cache_file=1';
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
   let rivaClient, request;
+  const sampleRate = 8000;
   try {
     rivaClient = await createRivaClient(riva_server_uri);
     request = new SynthesizeSpeechRequest();
     request.setVoiceName(voice);
     request.setLanguageCode(language);
-    request.setSampleRateHz(8000);
+    request.setSampleRateHz(sampleRate);
     request.setEncoding(AudioEncoding.LINEAR_PCM);
     request.setText(text);
   } catch (err) {
@@ -731,7 +757,11 @@ const synthNvidia = async(client, logger, {credentials, stats, language,  voice,
         logger.info({err, voice, language}, 'error synthesizing speech using Nvidia');
         return reject(err);
       }
-      resolve(Buffer.from(response.getAudio()));
+      resolve({
+        audioContent: Buffer.from(response.getAudio()),
+        extension: 'r8',
+        sampleRate
+      });
     });
   });
 };
@@ -753,14 +783,13 @@ const synthCustomVendor = async(logger, {credentials, stats, language, voice, te
       text
     });
 
-    const regex = /\.[^\.]*$/g;
     const mime = response.headers['content-type'];
     const buffer = await response.arrayBuffer();
-    const fileExtension = getFileExtFromMime(mime);
+    const [extension, sampleRate] = getFileExtFromMime(mime);
     return {
-      audioBuffer: buffer,
-      filePath: filePath.replace(regex, fileExtension),
-      fileExtension
+      audioContent: buffer,
+      extension,
+      sampleRate
     };
   } catch (err) {
     logger.info({err}, `Vendor ${vendor} returned error`);
@@ -786,9 +815,13 @@ const synthElevenlabs = async(logger, {
     if (opts.voice_settings?.similarity_boost) params += `,similarity_boost=${opts.voice_settings.similarity_boost}`;
     if (opts.voice_settings?.stability) params += `,stability=${opts.voice_settings.stability}`;
     if (opts.voice_settings?.style) params += `,style=${opts.voice_settings.style}`;
+    if (opts.voice_settings?.speed !== null && opts.voice_settings?.speed !== undefined)
+      params += `,speed=${opts.voice_settings.speed}`;
     if (opts.voice_settings?.use_speaker_boost === false) params += ',use_speaker_boost=false';
     if (opts.previous_text) params += `,previous_text=${opts.previous_text}`;
     if (opts.next_text) params += `,next_text=${opts.next_text}`;
+    if (opts.pronunciation_dictionary_locators && Array.isArray(opts.pronunciation_dictionary_locators))
+      params += `,pronunciation_dictionary_locators=${JSON.stringify(opts.pronunciation_dictionary_locators)}`;
     params += '}';
 
     return {
@@ -806,7 +839,7 @@ const synthElevenlabs = async(logger, {
       'Accept': 'audio/mpeg',
       'Content-Type': 'application/json'
     });
-    const mp3 = await post(`/v1/text-to-speech/${voice}${optimize_streaming_latency}`, {
+    const audioContent = await post(`/v1/text-to-speech/${voice}${optimize_streaming_latency}`, {
       text,
       model_id,
       voice_settings: {
@@ -815,7 +848,11 @@ const synthElevenlabs = async(logger, {
       },
       ...opts
     });
-    return mp3;
+    return {
+      audioContent,
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, 'synth Elevenlabs returned error');
     stats.increment('tts.count', ['vendor:elevenlabs', 'accepted:no']);
@@ -826,11 +863,10 @@ const synthElevenlabs = async(logger, {
 const synthPlayHT = async(client, logger, {
   credentials, options, stats, voice, language, text, renderForCaching, disableTtsStreaming
 }) => {
-  const {api_key, user_id, voice_engine, options: credOpts} = credentials;
+  const {api_key, user_id, voice_engine, playht_tts_uri, options: credOpts} = credentials;
   const opts = !!options && Object.keys(options).length !== 0 ? options : JSON.parse(credOpts || '{}');
 
-  let synthesizeUrl = 'https://api.play.ht/api/v2/tts/stream';
-
+  let synthesizeUrl = playht_tts_uri ? `${playht_tts_uri}/api/v2/tts/stream` : 'https://api.play.ht/api/v2/tts/stream';
   // If model is play3.0, the synthesizeUrl is got from authentication endpoint
   if (voice_engine === 'Play3.0') {
     try {
@@ -897,7 +933,7 @@ const synthPlayHT = async(client, logger, {
       'Content-Type': 'application/json'
     });
 
-    const mp3 = await post(synthesizeUrl, {
+    const audioContent = await post(synthesizeUrl, {
       text,
       ...(voice_engine === 'Play3.0' && { language }),
       voice,
@@ -906,7 +942,11 @@ const synthPlayHT = async(client, logger, {
       sample_rate: 8000,
       ...opts
     });
-    return mp3;
+    return {
+      audioContent,
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, 'synth PlayHT returned error');
     stats.increment('tts.count', ['vendor:playht', 'accepted:no']);
@@ -915,7 +955,7 @@ const synthPlayHT = async(client, logger, {
 };
 
 const synthRimelabs = async(logger, {
-  credentials, options, stats, voice, text, renderForCaching, disableTtsStreaming
+  credentials, options, stats, language, voice, text, renderForCaching, disableTtsStreaming
 }) => {
   const {api_key, model_id, options: credOpts} = credentials;
   const opts = !!options && Object.keys(options).length !== 0 ? options : JSON.parse(credOpts || '{}');
@@ -926,10 +966,16 @@ const synthRimelabs = async(logger, {
     params += `{api_key=${api_key}`;
     params += `,model_id=${model_id}`;
     params += ',vendor=rimelabs';
+    params += `,language=${language}`;
     params += `,voice=${voice}`;
     params += ',write_cache_file=1';
     if (opts.speedAlpha) params += `,speed_alpha=${opts.speedAlpha}`;
     if (opts.reduceLatency) params += `,reduce_latency=${opts.reduceLatency}`;
+    // Arcana model parameters
+    if (opts.temperature) params += `,temperature=${opts.temperature}`;
+    if (opts.repetition_penalty) params += `,repetition_penalty=${opts.repetition_penalty}`;
+    if (opts.top_p) params += `,top_p=${opts.top_p}`;
+    if (opts.max_tokens) params += `,max_tokens=${opts.max_tokens}`;
     params += '}';
 
     return {
@@ -945,14 +991,20 @@ const synthRimelabs = async(logger, {
       'Accept': 'audio/mp3',
       'Content-Type': 'application/json'
     });
-    const mp3 = await post('/v1/rime-tts', {
+    const sampleRate = 8000;
+    const audioContent = await post('/v1/rime-tts', {
       speaker: voice,
       text,
       modelId: model_id,
-      samplingRate: 8000,
+      samplingRate: sampleRate,
+      lang: language,
       ...opts
     });
-    return mp3;
+    return {
+      audioContent,
+      extension: 'mp3',
+      sampleRate
+    };
   } catch (err) {
     logger.info({err}, 'synth rimelabs returned error');
     stats.increment('tts.count', ['vendor:rimelabs', 'accepted:no']);
@@ -986,13 +1038,17 @@ const synthVerbio = async(client, logger, {credentials, stats, voice, text, rend
       'User-Agent': 'jambonz',
       'Content-Type': 'application/json'
     });
-    const r8 = await post('/api/v1/synthesize', {
+    const audioContent = await post('/api/v1/synthesize', {
       voice_id: voice,
       output_sample_rate: '8k',
       output_encoding: 'pcm16',
       text
     });
-    return r8;
+    return {
+      audioContent,
+      extension: 'r8',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, 'synth Verbio returned error');
     stats.increment('tts.count', ['vendor:verbio', 'accepted:no']);
@@ -1000,7 +1056,8 @@ const synthVerbio = async(client, logger, {credentials, stats, voice, text, rend
   }
 };
 
-const synthWhisper = async(logger, {credentials, stats, voice, text, renderForCaching, disableTtsStreaming}) => {
+const synthWhisper = async(logger, {credentials, stats, voice, text, instructions,
+  renderForCaching, disableTtsStreaming}) => {
   const {api_key, model_id, baseURL, timeout, speed} = credentials;
   /* if the env is set to stream then bag out, unless we are specifically rendering to generate a cache file */
   if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
@@ -1011,6 +1068,8 @@ const synthWhisper = async(logger, {credentials, stats, voice, text, renderForCa
     params += `,voice=${voice}`;
     params += ',write_cache_file=1';
     if (speed) params += `,speed=${speed}`;
+    // comma is used to separated parameters in freeswitch tts module
+    if (instructions) params += `,instructions=${instructions.replace(/\n/g, ' ').replace(/,/g, ';')}`;
     params += '}';
 
     return {
@@ -1030,9 +1089,14 @@ const synthWhisper = async(logger, {credentials, stats, voice, text, renderForCa
       model: model_id,
       voice,
       input: text,
+      ...(instructions && {instructions}),
       response_format: 'mp3'
     });
-    return Buffer.from(await mp3.arrayBuffer());
+    return {
+      audioContent: Buffer.from(await mp3.arrayBuffer()),
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, 'synth whisper returned error');
     stats.increment('tts.count', ['vendor:openai', 'accepted:no']);
@@ -1064,10 +1128,14 @@ const synthDeepgram = async(logger, {credentials, stats, model, text, renderForC
       'Accept': 'audio/mpeg',
       'Content-Type': 'application/json'
     });
-    const mp3 = await post(`/v1/speak?model=${model}`, {
+    const audioContent = await post(`/v1/speak?model=${model}`, {
       text
     });
-    return mp3;
+    return {
+      audioContent,
+      extension: 'mp3',
+      sampleRate: 8000
+    };
   } catch (err) {
     logger.info({err}, 'synth Deepgram returned error');
     stats.increment('tts.count', ['vendor:deepgram', 'accepted:no']);
@@ -1104,6 +1172,7 @@ const synthCartesia = async(logger, {
 
   try {
     const client = new CartesiaClient({ apiKey: api_key });
+    const sampleRate = 48000;
     const mp3 = await client.tts.bytes({
       modelId: model_id,
       transcript: text,
@@ -1119,7 +1188,7 @@ const synthCartesia = async(logger, {
         ),
         ...(opts.speed || opts.emotion && {
           experimentalControls: {
-            ...(opts.speed && {speed: opts.speed}),
+            ...(opts.speed !== null && opts.speed !== undefined && {speed: opts.speed}),
             ...(opts.emotion && {emotion: opts.emotion}),
           }
         })
@@ -1128,10 +1197,14 @@ const synthCartesia = async(logger, {
       outputFormat: {
         container: 'mp3',
         bitRate: 128000,
-        sampleRate: 8000
+        sampleRate
       },
     });
-    return Buffer.from(mp3);
+    return {
+      audioContent: Buffer.from(mp3),
+      extension: 'mp3',
+      sampleRate
+    };
   } catch (err) {
     logger.info({err}, 'synth Cartesia returned error');
     stats.increment('tts.count', ['vendor:cartesia', 'accepted:no']);
@@ -1144,22 +1217,22 @@ const getFileExtFromMime = (mime) => {
   switch (mime) {
     case 'audio/wav':
     case 'audio/x-wav':
-      return '.wav';
+      return ['wav', 8000];
     case /audio\/l16.*rate=8000/.test(mime) ? mime : 'cant match value':
-      return '.r8';
+      return ['r8', 8000];
     case /audio\/l16.*rate=16000/.test(mime) ? mime : 'cant match value':
-      return '.r16';
+      return ['r16', 16000];
     case /audio\/l16.*rate=24000/.test(mime) ? mime : 'cant match value':
-      return '.r24';
+      return ['r24', 24000];
     case /audio\/l16.*rate=32000/.test(mime) ? mime : 'cant match value':
-      return '.r32';
+      return ['r32', 32000];
     case /audio\/l16.*rate=48000/.test(mime) ? mime : 'cant match value':
-      return '.r48';
+      return ['r48', 48000];
     case 'audio/mpeg':
     case 'audio/mp3':
-      return '.mp3';
+      return ['mp3', 8000];
     default:
-      return '.wav';
+      return ['wav', 8000];
   }
 };
 

lib/utils.js

@@ -6,7 +6,7 @@ const pool = new Pool('https://auth.crt.nuance.com');
 const NUANCE_AUTH_ENDPOINT = 'tts.api.nuance.com:443';
 const grpc = require('@grpc/grpc-js');
 const formurlencoded = require('form-urlencoded');
-const { JAMBONES_DISABLE_TTS_STREAMING, JAMBONES_TTS_TRIM_SILENCE, TMP_FOLDER, HTTP_TIMEOUT } = require('./config');
+const { TMP_FOLDER, HTTP_TIMEOUT } = require('./config');
 
 const debug = require('debug')('jambonz:realtimedb-helpers');
 /**
@@ -17,68 +17,27 @@ const debug = require('debug')('jambonz:realtimedb-helpers');
 //const nuanceClientMap = new Map();
 
 function makeSynthKey({
-  account_sid = '', vendor, language, voice, engine = '', text,
-  renderForCaching = false}) {
+  account_sid = '',
+  vendor,
+  language,
+  voice,
+  engine = '',
+  model = '',
+  text,
+  instructions = '',
+}) {
   const hash = crypto.createHash('sha1');
-  hash.update(`${language}:${vendor}:${voice}:${engine}:${text}`);
+  hash.update(`${language}:${vendor}:${voice}:${engine}:${model}:${text}:${instructions}`);
   const hexHashKey = hash.digest('hex');
   const accountKey = account_sid ? `:${account_sid}` : '';
-  const namespace = vendor.startsWith('custom') ? vendor : getFileExtension({vendor, voice, renderForCaching});
-  const key = `tts${accountKey}:${namespace}:${hexHashKey}`;
+  const key = `tts${accountKey}:${hexHashKey}`;
   return key;
 }
 
-function makeFilePath({vendor, voice, key, salt = '', renderForCaching = false}) {
-  const extension = getFileExtension({vendor, renderForCaching, voice});
+function makeFilePath({key, salt = '', extension}) {
   return `${TMP_FOLDER}/${key.replace('tts:', `tts-${salt}`)}.${extension}`;
 }
 
-function getFileExtension({vendor, voice, renderForCaching = false}) {
-  const mp3Extension = 'mp3';
-  const r8Extension = 'r8';
-  const wavExtension = 'wav';
-
-  switch (vendor) {
-    case 'azure':
-    case 'microsoft':
-      if (!renderForCaching && !JAMBONES_DISABLE_TTS_STREAMING || JAMBONES_TTS_TRIM_SILENCE) {
-        return r8Extension;
-      } else {
-        return mp3Extension;
-      }
-    case 'deepgram':
-    case 'elevenlabs':
-    case 'rimelabs':
-    case 'playht':
-    case 'cartesia':
-      if (renderForCaching || JAMBONES_DISABLE_TTS_STREAMING) {
-        return mp3Extension;
-      } else {
-        return r8Extension;
-      }
-    case 'nuance':
-    case 'nvidia':
-    case 'verbio':
-      return r8Extension;
-    case 'google':
-      // google voice cloning just support wav.
-      if (typeof voice === 'object' && voice.voice_cloning_key) {
-        return wavExtension;
-      } else {
-        return mp3Extension;
-      }
-    default:
-      // If vendor is custom
-      if (vendor.startsWith('custom')) {
-        if (renderForCaching || JAMBONES_DISABLE_TTS_STREAMING) {
-          return mp3Extension;
-        } else {
-          return r8Extension;
-        }
-      }
-      return mp3Extension;
-  }
-}
 
 const noopLogger = {
   info: () => {},

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jambonz/speech-utils",
-  "version": "0.1.24",
+  "version": "0.2.11",
   "description": "TTS-related speech utilities for jambonz",
   "main": "index.js",
   "author": "Dave Horton",
@@ -26,6 +26,7 @@
   },
   "homepage": "https://github.com/jambonz/speech-utils#readme",
   "dependencies": {
+    "23": "^0.0.0",
     "@aws-sdk/client-polly": "^3.496.0",
     "@aws-sdk/client-sts": "^3.496.0",
     "@cartesia/cartesia-js": "^2.1.0",
@@ -38,8 +39,8 @@
     "google-protobuf": "^3.21.2",
     "ibm-watson": "^8.0.0",
     "microsoft-cognitiveservices-speech-sdk": "1.38.0",
-    "openai": "^4.25.0",
-    "undici": "^6.4.0"
+    "openai": "^4.98.0",
+    "undici": "^7.5.0"
   },
   "devDependencies": {
     "config": "^3.3.11",

test/synth.js

@@ -5,7 +5,7 @@ const fs = require('fs');
 const {makeSynthKey} = require('../lib/utils');
 const logger = require('pino')();
 const bent = require('bent');
-const getJSON = bent('json')
+const getJSON = bent('json');
 
 process.on('unhandledRejection', (reason, p) => {
   console.log('Unhandled Rejection at: Promise', p, 'reason:', reason);
@@ -720,7 +720,7 @@ test('Cartesia speech synth tests', async(t) => {
   client.quit();
 });
 
-test('rimelabs speech synth tests', async(t) => {
+test('rimelabs speech synth tests mist', async(t) => {
   const fn = require('..');
   const {synthAudio, client} = fn(opts, logger);
 
@@ -730,7 +730,7 @@ test('rimelabs speech synth tests', async(t) => {
   }
   const text = 'Hi there and welcome to jambones!';
   try {
-    let opts = await synthAudio(stats, {
+    const opts = await synthAudio(stats, {
       vendor: 'rimelabs',
       credentials: {
         api_key: process.env.RIMELABS_API_KEY,
@@ -740,7 +740,7 @@ test('rimelabs speech synth tests', async(t) => {
           reduceLatency: false
         })
       },
-      language: 'en-US',
+      language: 'eng',
       voice: 'amber',
       text,
       renderForCaching: true
@@ -754,6 +754,40 @@ test('rimelabs speech synth tests', async(t) => {
   client.quit();
 });
 
+test('rimelabs speech synth tests mistv2', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.RIMELABS_API_KEY) {
+    t.pass('skipping rimelabs speech synth tests since RIMELABS_API_KEY is not provided');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambones!';
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'rimelabs',
+      credentials: {
+        api_key: process.env.RIMELABS_API_KEY,
+        model_id: 'mistv2',
+        options: JSON.stringify({
+          speedAlpha: 1.0,
+          reduceLatency: false
+        })
+      },
+      language: 'spa',
+      voice: 'pablo',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache, `successfully synthesized rimelabs mistv2 audio to ${opts.filePath}`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
 test('whisper speech synth tests', async(t) => {
   const fn = require('..');
   const {synthAudio, client} = fn(opts, logger);
@@ -853,11 +887,12 @@ test('TTS Cache tests', async(t) => {
     // save some random tts keys to cache
     const minRecords = 8;
     for (const i in Array(minRecords).fill(0)) {
-      await client.set(makeSynthKey({vendor: i, language: i, voice: i, engine: i, text: i}), i);
+      await client.set(makeSynthKey({vendor: i, language: i, voice: i, engine: i, model: i, text: i,
+        instructions: i}), i);
     }
     const count = await getTtsSize();
     t.ok(count >= minRecords, 'getTtsSize worked.');
-    
+
     const {purgedCount} = await purgeTtsCache();
     t.ok(purgedCount >= minRecords, `successfully purged at least ${minRecords} tts records from cache`);
 
@@ -872,7 +907,7 @@ test('TTS Cache tests', async(t) => {
   try {
     // save some random tts keys to cache
     for (const i in Array(10).fill(0)) {
-      await client.set(makeSynthKey({vendor: i, language: i, voice: i, engine: i, text: i}), i);
+      await client.set(makeSynthKey({vendor: i, language: i, voice: i, engine: i, text: i, instructions: i}), i);
     }
     // save a specific key to tts cache
     const opts = {vendor: 'aws', language: 'en-US', voice: 'MALE', engine: 'Engine', text: 'Hello World!'};
@@ -888,24 +923,15 @@ test('TTS Cache tests', async(t) => {
       language: 'non-existing',
       voice: 'non-existing',
     });
-    t.ok(purgedCountWhenErrored === 0, `purged no records when specified key was not found`);
-    t.ok(error, `error returned when specified key was not found`);
+    t.ok(purgedCountWhenErrored === 0, 'purged no records when specified key was not found');
+    t.ok(error, 'error returned when specified key was not found');
 
     // make sure other tts keys are still there
-    const cached = await client.keys('tts:*')
-    t.ok(cached.length >= 1, `successfully kept all non-specified tts records in cache`);
+    const cached = await client.keys('tts:*');
+    t.ok(cached.length >= 1, 'successfully kept all non-specified tts records in cache');
 
-    // retrieve keys from cache and check the key contains the file extension    
-    let key = cached[0];
-    t.ok(key.includes('mp3'), `tts cache extension shoult be part of the key and equal mp3`);
-
-    process.env.VG_TRIM_TTS_SILENCE = 'true';    
+    process.env.VG_TRIM_TTS_SILENCE = 'true';
     await client.set(makeSynthKey({ vendor: 'azure' }), 'value');
-
-    const r8Keys = await client.keys('tts:r8*');
-    key = r8Keys[0];
-    t.ok(key.includes('r8'), `tts cache extension shoult be part of the key and equal r8`);
-
   } catch (err) {
     console.error(JSON.stringify(err));
     t.end(err);
@@ -919,10 +945,10 @@ test('TTS Cache tests', async(t) => {
     const account_sid = "12412512_cabc_5aff"
     const account_sid2 = "22412512_cabc_5aff"
     for (const i in Array(minRecords).fill(0)) {
-      await client.set(makeSynthKey({account_sid, vendor: i, language: i, voice: i, engine: i, text: i}), i);
+      await client.set(makeSynthKey({account_sid, vendor: i, language: i, voice: i, engine: i, text: i, instructions: i}), i);
     }
     for (const i in Array(minRecords).fill(0)) {
-      await client.set(makeSynthKey({account_sid: account_sid2, vendor: i, language: i, voice: i, engine: i, text: i}), i);
+      await client.set(makeSynthKey({account_sid: account_sid2, vendor: i, language: i, voice: i, engine: i, text: i, instructions: 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}`);