@jambonz/speech-utils 0.0.19 → 0.0.20

package/lib/synth-audio.js

@@ -14,7 +14,13 @@ const {
   CancellationDetails,
   SpeechSynthesisOutputFormat
 } = sdk;
-const {makeSynthKey, createNuanceClient, createKryptonClient, createRivaClient, noopLogger} = require('./utils');
+const {
+  makeSynthKey,
+  createNuanceClient,
+  createKryptonClient,
+  createRivaClient,
+  noopLogger
+} = require('./utils');
 const getNuanceAccessToken = require('./get-nuance-access-token');
 const {
   SynthesisRequest,
@@ -297,6 +303,47 @@ const synthIbm = async(logger, {credentials, stats, voice, text}) => {
   }
 };
 
+async function _synthOnPremMicrosoft(logger, {
+  credentials,
+  stats,
+  language,
+  voice,
+  text,
+  filePath
+}) {
+  const {use_custom_tts, custom_tts_endpoint_url} = 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
+     * Otherwise sending plain text we get "Voice does not match"
+     */
+    content = `<speak>${text}</speak>`;
+  }
+
+  if (content.startsWith('<speak>')) {
+    /* microsoft enforces some properties and uses voice xml element so if the user did not supply do it for them */
+    const words = content.slice(7, -8).trim().replace(/(\r\n|\n|\r)/gm, ' ');
+    // eslint-disable-next-line max-len
+    content = `<speak version="1.0" xmlns="http://www.w3.org/2001/10/synthesis" xmlns:mstts="https://www.w3.org/2001/mstts" xml:lang="${language}"><voice name="${voice}">${words}</voice></speak>`;
+    logger.info({content}, 'synthMicrosoft');
+  }
+
+  try {
+    const post = bent('POST', 'buffer', {
+      'X-Microsoft-OutputFormat': 'audio-16khz-128kbitrate-mono-mp3',
+      'Content-Type': 'application/ssml+xml',
+      'User-Agent': 'Jambonz'
+    });
+    const mp3 = await post(custom_tts_endpoint_url, content);
+    return mp3;
+  } catch (err) {
+    logger.info({err}, '_synthMicrosoftByHttp returned error');
+    throw err;
+  }
+}
+
 const synthMicrosoft = async(logger, {
   credentials,
   stats,
@@ -306,7 +353,17 @@ const synthMicrosoft = async(logger, {
   filePath
 }) => {
   try {
-    const {api_key: apiKey, region, use_custom_tts, custom_tts_endpoint} = credentials;
+    const {api_key: apiKey, region, use_custom_tts, custom_tts_endpoint, custom_tts_endpoint_url} = credentials;
+    if (use_custom_tts && custom_tts_endpoint_url) {
+      return await _synthOnPremMicrosoft(logger, {
+        credentials,
+        stats,
+        language,
+        voice,
+        text,
+        filePath
+      });
+    }
     const trimSilence = filePath.endsWith('.r8');
     let content = text;
     const speechConfig = SpeechConfig.fromSubscription(apiKey, region);
@@ -314,12 +371,13 @@ const synthMicrosoft = async(logger, {
     speechConfig.speechSynthesisVoiceName = voice;
     if (use_custom_tts && custom_tts_endpoint) {
       speechConfig.endpointId = custom_tts_endpoint;
-
+    }
+    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"
        */
-      if (!content.startsWith('<speak')) content = `<speak>${text}</speak>`;
+      content = `<speak>${text}</speak>`;
     }
     speechConfig.speechSynthesisOutputFormat = trimSilence ?
       SpeechSynthesisOutputFormat.Raw8Khz16BitMonoPcm :

package/lib/utils.js

@@ -106,7 +106,6 @@ const createRivaClient = async(rivaUri) => {
   return client;
 };
 
-
 module.exports = {
   makeSynthKey,
   makeNuanceKey,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jambonz/speech-utils",
-  "version": "0.0.19",
+  "version": "0.0.20",
   "description": "TTS-related speech utilities for jambonz",
   "main": "index.js",
   "author": "Dave Horton",
@@ -32,7 +32,7 @@
     "form-urlencoded": "^6.1.0",
     "google-protobuf": "^3.21.2",
     "ibm-watson": "^8.0.0",
-    "microsoft-cognitiveservices-speech-sdk": "^1.26.0",
+    "microsoft-cognitiveservices-speech-sdk": "^1.31.0",
     "ioredis": "^5.3.2",
     "undici": "^5.21.0"
   },