1.0.18

feat: add fishaudio (Fish Audio) TTS support (#159 )
Adds synthFishaudio with both arms: the say: streaming url consumed by the mediajam dialect, and a POST /v1/tts cache render. The render asks for raw pcm at 8k and returns extension r8 because fish's wav output carries a placeholder RIFF size, the same problem gradium has. Fish is a voice-cloning vendor, so the voice is a reference_id; the sentinel 'default' means send none and use fish's own default voice. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
2026-10-03 23:33:59 +00:00 · 2026-08-23 13:12:37 -04:00 · 2026-08-23 13:11:59 -04:00 · 2026-08-16 12:19:59 -04:00 · 2026-08-10 16:45:32 -04:00 · 2026-08-10 16:45:07 -04:00
17 changed files with 996 additions and 8216 deletions
@@ -0,0 +1,84 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Overview
+
+`@jambonz/speech-utils` is a Node.js library providing TTS (Text-to-Speech) utilities for the jambonz CPaaS platform. It handles speech synthesis with caching through Redis and supports multiple TTS vendors.
+
+## Commands
+
+```bash
+# Run tests (requires Docker for Redis)
+npm test
+
+# Run linter
+npm run jslint
+
+# Auto-fix lint issues
+npm run jslint:fix
+
+# Generate coverage report
+npm run coverage
+```
+
+## Testing
+
+Tests use `tape` and require Redis. The test harness automatically starts/stops Redis via Docker Compose (`test/docker-compose-testbed.yaml`).
+
+Most tests are conditional based on environment variables for vendor credentials:
+- `GCP_FILE` or `GCP_JSON_KEY` - Google TTS
+- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`, `AWS_REGION` - AWS Polly
+- `MICROSOFT_API_KEY`, `MICROSOFT_REGION` - Azure TTS
+- `ELEVENLABS_API_KEY`, `ELEVENLABS_VOICE_ID`, `ELEVENLABS_MODEL_ID` - ElevenLabs
+- `OPENAI_API_KEY` - OpenAI Whisper TTS
+- And others per vendor
+
+Redis config is in `config/test.json` (port 3379).
+
+## Architecture
+
+### Entry Point
+
+`index.js` exports a factory function that takes Redis options and a logger, returning an object with these methods:
+- `synthAudio` - Main synthesis function
+- `getTtsVoices` - List available voices for a vendor
+- `purgeTtsCache` / `getTtsSize` / `addFileToCache` - Cache management
+- `getAwsAuthToken` - Token management
+
+### Core Module: `lib/synth-audio.js`
+
+The `synthAudio` function handles synthesis for all vendors. Key behaviors:
+1. **Cache check**: Generates SHA1 hash key from (vendor, language, voice, engine, model, text, instructions)
+2. **Streaming vs non-streaming**: When `JAMBONES_DISABLE_TTS_STREAMING` is not set and `renderForCaching=false`, returns `say:{params}text` format for FreeSWITCH streaming playback instead of generating files
+3. **Vendor dispatch**: Switch statement routes to vendor-specific synth functions (`synthGoogle`, `synthPolly`, `synthMicrosoft`, etc.)
+4. **Caching**: Stores audio as base64 JSON in Redis with configurable TTL (default 4 hours)
+
+### Supported Vendors
+
+google, aws/polly, microsoft/azure, nvidia (Riva), wellsaid, elevenlabs, cartesia, inworld, rimelabs, whisper (OpenAI), deepgram, resemble, custom:*
+
+### gRPC Stubs
+
+`stubs/riva/` contains generated protobuf/gRPC code for NVIDIA Riva.
+
+## Environment Variables
+
+Key configuration via env vars (see `lib/config.js`):
+- `JAMBONES_DISABLE_TTS_STREAMING` - Force non-streaming mode
+- `JAMBONES_DISABLE_AZURE_TTS_STREAMING` - Azure-specific streaming disable
+- `JAMBONES_TTS_CACHE_DURATION_MINS` - Cache TTL in minutes (default: 240)
+- `JAMBONES_TTS_TRIM_SILENCE` - Trim trailing silence from audio
+- `JAMBONES_TMP_FOLDER` - Temp folder for audio files (default: /tmp)
+- `JAMBONES_HTTP_PROXY_IP`, `JAMBONES_HTTP_PROXY_PORT` - HTTP proxy for Azure
+- `JAMBONES_AZURE_ENABLE_SSML` - Force SSML wrapper for Azure plain text
+
+## Key Dependencies
+
+- `@jambonz/realtimedb-helpers` - Redis client and hash utilities
+- `@google-cloud/text-to-speech` - Google TTS
+- `@aws-sdk/client-polly` - AWS Polly
+- `microsoft-cognitiveservices-speech-sdk` - Azure TTS
+- `@grpc/grpc-js` - gRPC for Riva
+- `openai` - OpenAI Whisper TTS
+- `bent` - HTTP client for REST-based vendors
@@ -1,11 +0,0 @@
-#!/bin/sh
-
-mkdir -p stubs/nuance
-
-for FILE in ./protos/nuance/*; do 
-  grpc_tools_node_protoc   \
-  --js_out=import_style=commonjs,binary:./stubs/nuance \
-  --grpc_out=grpc_js:./stubs/nuance \
-  --proto_path=./protos/nuance  \
-  $FILE
-done
@@ -14,8 +14,6 @@ module.exports = (opts, logger) => {
    purgeTtsCache: require('./lib/purge-tts-cache').bind(null, client, logger),
    addFileToCache: require('./lib/add-file-to-cache').bind(null, client, logger),
    synthAudio: require('./lib/synth-audio').bind(null, client, createHash, retrieveHash, logger),
-    getVerbioAccessToken: require('./lib/get-verbio-token').bind(null, client, logger),
-    getNuanceAccessToken: require('./lib/get-nuance-access-token').bind(null, client, logger),

    getAwsAuthToken: require('./lib/get-aws-sts-token').bind(null, logger, createHash, retrieveHash),
    getTtsVoices: require('./lib/get-tts-voices').bind(null, client, createHash, retrieveHash, logger),
@@ -1,49 +0,0 @@
-const formurlencoded = require('form-urlencoded');
-const {Pool} = require('undici');
-const pool = new Pool('https://auth.crt.nuance.com');
-const {makeNuanceKey, makeBasicAuthHeader, noopLogger} = require('./utils');
-const { HTTP_TIMEOUT } = require('./config');
-const debug = require('debug')('jambonz:realtimedb-helpers');
-
-async function getNuanceAccessToken(client, logger, clientId, secret, scope) {
-  logger = logger || noopLogger;
-  try {
-    const key = makeNuanceKey(clientId, secret, scope);
-    const access_token = await client.get(key);
-    if (access_token) return {access_token, servedFromCache: true};
-
-    /* access token not found in cache, so fetch it from Nuance */
-    const payload = {
-      grant_type: 'client_credentials',
-      scope
-    };
-    const auth = makeBasicAuthHeader(clientId, secret);
-    const {statusCode, headers, body} =  await pool.request({
-      path: '/oauth2/token',
-      method: 'POST',
-      headers: {
-        ...auth,
-        'Content-Type': 'application/x-www-form-urlencoded'
-      },
-      body: formurlencoded(payload),
-      timeout: HTTP_TIMEOUT,
-      followRedirects: false
-    });
-
-    if (200 !== statusCode) {
-      logger.debug({statusCode, headers, body: body.text()}, 'error fetching access token from Nuance');
-      const err = new Error();
-      err.statusCode = statusCode;
-      throw err;
-    }
-    const json = await body.json();
-    await client.set(key, json.access_token, 'EX', json.expires_in - 30);
-    return {...json, servedFromCache: false};
-  } catch (err) {
-    debug(err, `getNuanceAccessToken: Error retrieving Nuance access token for client_id ${clientId}`);
-    logger.error(err, `getNuanceAccessToken: Error retrieving Nuance access token for client_id ${clientId}`);
-    throw err;
-  }
-}
-
-module.exports = getNuanceAccessToken;
@@ -1,74 +1,8 @@
 const assert = require('assert');
-const {noopLogger, createNuanceClient, createKryptonClient} = require('./utils');
-const getNuanceAccessToken = require('./get-nuance-access-token');
-const getVerbioAccessToken = require('./get-verbio-token');
-const {GetVoicesRequest, Voice} = require('../stubs/nuance/synthesizer_pb');
+const {noopLogger} = require('./utils');
 const ttsGoogle = require('@google-cloud/text-to-speech');
 const { PollyClient, DescribeVoicesCommand } = require('@aws-sdk/client-polly');
 const getAwsAuthToken = require('./get-aws-sts-token');
-const {Pool} = require('undici');
-const { HTTP_TIMEOUT } = require('./config');
-const verbioVoicePool = new Pool('https://us.rest.speechcenter.verbio.com');
-
-const getNuanceVoices = async(client, logger, credentials) => {
-  const {client_id: clientId, secret: secret, nuance_tts_uri} = credentials;
-
-  return new Promise(async(resolve, reject) => {
-    /* get a nuance access token */
-    let token, nuanceClient;
-    try {
-      if (nuance_tts_uri) {
-        nuanceClient = await createKryptonClient(nuance_tts_uri);
-      }
-      else {
-        const access_token = await getNuanceAccessToken(client, logger, clientId, secret, 'tts');
-        token = access_token.access_token;
-        nuanceClient = await createNuanceClient(token);
-      }
-    } catch (err) {
-      logger.error({err}, 'getTtsVoices: error retrieving access token');
-      return reject(err);
-    }
-    /* retrieve all voices */
-    const v = new Voice();
-    const request = new GetVoicesRequest();
-    request.setVoice(v);
-
-    nuanceClient.getVoices(request, (err, response) => {
-      if (err) {
-        logger.error({err, clientId, secret, token}, 'getTtsVoices: error retrieving voices');
-        return reject(err);
-      }
-
-      /* return all the voices that are not restricted and eliminate duplicates */
-      const voices = response.getVoicesList()
-        .map((v) => {
-          return {
-            language: v.getLanguage(),
-            name: v.getName(),
-            model: v.getModel(),
-            gender: v.getGender() === 1 ? 'male' : 'female',
-            restricted: v.getRestricted()
-          };
-        });
-      const v = voices
-        .filter((v) => v.restricted === false)
-        .map((v) => {
-          delete v.restricted;
-          return v;
-        })
-        .sort((a, b) => {
-          if (a.language < b.language) return -1;
-          if (a.language > b.language) return 1;
-          if (a.name < b.name) return -1;
-          return 1;
-        });
-      const arr = [...new Set(v.map((v) => JSON.stringify(v)))]
-        .map((v) => JSON.parse(v));
-      resolve(arr);
-    });
-  });
-};

 const getGoogleVoices = async(_client, logger, credentials) => {
  const client = new ttsGoogle.TextToSpeechClient({credentials});
@@ -109,26 +43,6 @@ const getAwsVoices = async(_client, createHash, retrieveHash, logger, credential
  }
 };

-const getVerbioVoices = async(client, logger, credentials) => {
-  try {
-    const access_token = await getVerbioAccessToken(client, logger, credentials);
-    const { body} =  await verbioVoicePool.request({
-      path: '/api/v1/voices',
-      method: 'GET',
-      headers: {
-        'Authorization': `Bearer ${access_token.access_token}`,
-        'User-Agent': 'jambonz'
-      },
-      timeout: HTTP_TIMEOUT,
-      followRedirects: false
-    });
-    return await body.json();
-  } catch (err) {
-    logger.info({err}, 'getVerbioVoices - failed to list voices for Verbio');
-    throw err;
-  }
-};
-
 /**
 * Synthesize speech to an mp3 file, and also cache the generated speech
 * in redis (base64 format) for 24 hours so as to avoid unnecessarily paying
@@ -148,19 +62,15 @@ const getVerbioVoices = async(client, logger, credentials) => {
 async function getTtsVoices(client, createHash, retrieveHash, logger, {vendor, credentials}) {
  logger = logger || noopLogger;

-  assert.ok(['nuance', 'google', 'aws', 'polly', 'verbio'].includes(vendor),
+  assert.ok(['google', 'aws', 'polly'].includes(vendor),
    `getTtsVoices not supported for vendor ${vendor}`);

  switch (vendor) {
-    case 'nuance':
-      return getNuanceVoices(client, logger, credentials);
    case 'google':
      return getGoogleVoices(client, logger, credentials);
    case 'aws':
    case 'polly':
      return getAwsVoices(client, createHash, retrieveHash, logger, credentials);
-    case 'verbio':
-      return getVerbioVoices(client, logger, credentials);
    default:
      break;
  }
@@ -1,51 +0,0 @@
-const {Pool} = require('undici');
-const { noopLogger, makeVerbioKey } = require('./utils');
-const { HTTP_TIMEOUT } = require('./config');
-const pool = new Pool('https://auth.speechcenter.verbio.com:444');
-const debug = require('debug')('jambonz:realtimedb-helpers');
-
-async function getVerbioAccessToken(client, logger, credentials) {
-  logger = logger || noopLogger;
-  const { client_id, client_secret } = credentials;
-  try {
-    const key = makeVerbioKey(client_id);
-    const access_token = await client.get(key);
-    if (access_token) {
-      return {access_token, servedFromCache: true};
-    }
-
-    const payload = {
-      client_id,
-      client_secret
-    };
-
-    const {statusCode, headers, body} =  await pool.request({
-      path: '/api/v1/token',
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        'User-Agent': 'jambonz'
-      },
-      body: JSON.stringify(payload),
-      timeout: HTTP_TIMEOUT,
-      followRedirects: false
-    });
-
-    if (200 !== statusCode) {
-      logger.debug({statusCode, headers, body: await body.text()}, 'error fetching access token from Verbio');
-      const err = new Error();
-      err.statusCode = statusCode;
-      throw err;
-    }
-    const json = await body.json();
-    const expiry =  Math.floor(json.expiration_time - Date.now() / 1000 - 30);
-    await client.set(key, json.access_token, 'EX', expiry);
-    return {...json, servedFromCache: false};
-  } catch (err) {
-    debug(err, `getVerbioAccessToken: Error retrieving Verbio access token for client_id ${client_id}`);
-    logger.error(err, `getVerbioAccessToken: Error retrieving Verbio access token for client_id ${client_id}`);
-    throw err;
-  }
-}
-
-module.exports = getVerbioAccessToken;
@@ -16,26 +16,10 @@ const {
 } = sdk;
 const {
  makeSynthKey,
-  createNuanceClient,
-  createKryptonClient,
  createRivaClient,
  noopLogger,
-  makeFilePath,
-  makePlayhtKey
+  makeFilePath
 } = require('./utils');
-const getNuanceAccessToken = require('./get-nuance-access-token');
-const getVerbioAccessToken = require('./get-verbio-token');
-const {
-  SynthesisRequest,
-  Voice,
-  AudioFormat,
-  AudioParameters,
-  PCM,
-  Input,
-  Text,
-  SSML,
-  EventParameters
-} = require('../stubs/nuance/synthesizer_pb');
 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');
@@ -95,10 +79,12 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
  let rtt;
  logger = logger || noopLogger;

-  assert.ok(['google', 'aws', 'polly', 'microsoft', 'wellsaid', 'nuance', 'nvidia', 'elevenlabs',
-    'whisper', 'deepgram', 'playht', 'rimelabs', 'verbio', 'cartesia', 'inworld', 'resemble'].includes(vendor) ||
+  assert.ok(['google', 'aws', 'polly', 'microsoft', 'wellsaid', 'nvidia', 'elevenlabs',
+    'whisper', 'deepgram', 'deepgramflux', 'rimelabs', 'cartesia', 'gradium', 'nineninesix', 'inworld', 'resemble',
+    'murf', 'xai', 'fishaudio']
+    .includes(vendor) ||
  vendor.startsWith('custom'),
-  `synthAudio supported vendors are google, aws, microsoft, nuance, nvidia and wellsaid ..etc, not ${vendor}`);
+  `synthAudio supported vendors are google, aws, microsoft, nvidia and wellsaid ..etc, not ${vendor}`);
  if ('google' === vendor) {
    assert.ok(language, 'synthAudio requires language when google is used');
  }
@@ -109,13 +95,6 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
    assert.ok(language || deploymentId, 'synthAudio requires language when microsoft is used');
    assert.ok(voice || deploymentId, 'synthAudio requires voice when microsoft is used');
  }
-  else if ('nuance' === vendor) {
-    assert.ok(voice, 'synthAudio requires voice when nuance is used');
-    if (!credentials.nuance_tts_uri) {
-      assert.ok(credentials.client_id, 'synthAudio requires client_id in credentials when nuance is used');
-      assert.ok(credentials.secret, 'synthAudio requires client_id in credentials when nuance is used');
-    }
-  }
  else if ('nvidia' === vendor) {
    assert.ok(voice, 'synthAudio requires voice when nvidia is used');
    assert.ok(language, 'synthAudio requires language when nvidia is used');
@@ -129,11 +108,6 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
    assert.ok(voice, 'synthAudio requires voice when elevenlabs is used');
    assert.ok(credentials.api_key, 'synthAudio requires api_key when elevenlabs is used');
    assert.ok(credentials.model_id, 'synthAudio requires model_id when elevenlabs is used');
-  } else if ('playht' === vendor) {
-    assert.ok(voice, 'synthAudio requires voice when playht is used');
-    assert.ok(credentials.api_key, 'synthAudio requires api_key when playht is used');
-    assert.ok(credentials.user_id, 'synthAudio requires user_id when playht is used');
-    assert.ok(credentials.voice_engine, 'synthAudio requires voice_engine when playht is used');
  } else if ('inworld' === vendor) {
    assert.ok(voice, 'synthAudio requires voice when inworld is used');
    assert.ok(credentials.api_key, 'synthAudio requires api_key when inworld is used');
@@ -148,20 +122,37 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
    assert.ok(credentials.api_key, 'synthAudio requires api_key when whisper is used');
  } else  if (vendor.startsWith('custom')) {
    assert.ok(credentials.custom_tts_url, `synthAudio requires custom_tts_url in credentials when ${vendor} is used`);
-  } else if ('verbio' === vendor) {
-    assert.ok(voice, 'synthAudio requires voice when verbio is used');
-    assert.ok(credentials.client_id, 'synthAudio requires client_id when verbio is used');
-    assert.ok(credentials.client_secret, 'synthAudio requires client_secret when verbio is used');
  } else if ('deepgram' === vendor) {
    if (!credentials.deepgram_tts_uri) {
      assert.ok(credentials.api_key, 'synthAudio requires api_key when deepgram is used');
    }
+  } else if ('deepgramflux' === vendor) {
+    // Deepgram Flux TTS (/v2/speak); the flux model rides on `voice`/`model`
+    if (!credentials.deepgram_tts_uri) {
+      assert.ok(credentials.api_key, 'synthAudio requires api_key when deepgramflux is used');
+    }
+  } else if ('xai' === vendor) {
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when xai is used');
  } else if ('cartesia' === vendor) {
    assert.ok(credentials.api_key, 'synthAudio requires api_key when cartesia is used');
    assert.ok(credentials.model_id, 'synthAudio requires model_id when cartesia is used');
+  } else if ('gradium' === vendor) {
+    assert.ok(voice, 'synthAudio requires voice when gradium is used');
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when gradium is used');
+  } else if ('nineninesix' === vendor) {
+    assert.ok(voice, 'synthAudio requires voice when nineninesix is used');
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when nineninesix is used');
+    assert.ok(credentials.model_id, 'synthAudio requires model_id when nineninesix is used');
+  } else if ('murf' === vendor) {
+    assert.ok(voice, 'synthAudio requires voice when murf is used');
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when murf is used');
  } else if (vendor === 'resemble') {
    assert.ok(voice, 'synthAudio requires voice when resemble is used');
    assert.ok(credentials.api_key, 'synthAudio requires api_key when resemble is used');
+  } else if ('fishaudio' === vendor) {
+    /* no voice assert: fish synthesizes with its own default voice when
+       reference_id is omitted, which is what the 'default' selection means */
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when fishaudio is used');
  }

  const key = makeSynthKey({
@@ -216,10 +207,6 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
        audioData = await synthMicrosoft(logger, {credentials, stats, language, voice, key, text, deploymentId,
          renderForCaching, disableTtsStreaming, disableTtsCache});
        break;
-      case 'nuance':
-        model = model || 'enhanced';
-        audioData = await synthNuance(client, logger, {credentials, stats, voice, model, key, text});
-        break;
      case 'nvidia':
        audioData = await synthNvidia(client, logger, {credentials, stats, language, voice, model, key, text,
          renderForCaching, disableTtsStreaming, disableTtsCache});
@@ -232,16 +219,26 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
          disableTtsCache});
        break;
-      case 'playht':
-        audioData = await synthPlayHT(client, logger, {
-          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
-          disableTtsCache});
-        break;
      case 'cartesia':
        audioData = await synthCartesia(logger, {
          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
          disableTtsCache});
        break;
+      case 'fishaudio':
+        audioData = await synthFishaudio(logger, {
+          credentials, options, stats, voice, key, text, renderForCaching, disableTtsStreaming,
+          disableTtsCache});
+        break;
+      case 'gradium':
+        audioData = await synthGradium(logger, {
+          credentials, options, stats, voice, key, text, renderForCaching, disableTtsStreaming,
+          disableTtsCache});
+        break;
+      case 'nineninesix':
+        audioData = await synthNineninesix(logger, {
+          credentials, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
+          disableTtsCache});
+        break;
      case 'inworld':
        audioData = await synthInworld(logger, {
          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
@@ -252,20 +249,29 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
          disableTtsCache});
        break;
+      case 'murf':
+        audioData = await synthMurf(logger, {
+          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
+          disableTtsCache});
+        break;
      case 'whisper':
        audioData = await synthWhisper(logger, {
          credentials, stats, voice, key, text, instructions, renderForCaching, disableTtsStreaming,
          disableTtsCache});
        break;
-      case 'verbio':
-        audioData = await synthVerbio(client, logger, {
-          credentials, stats, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache});
-        if (audioData?.filePath) return audioData;
-        break;
      case 'deepgram':
        audioData = await synthDeepgram(logger, {credentials, stats, model, key, text,
          renderForCaching, disableTtsStreaming, disableTtsCache});
        break;
+      case 'deepgramflux':
+        audioData = await synthDeepgramFlux(logger, {credentials, stats, model: model || voice, key, text,
+          renderForCaching, disableTtsStreaming, disableTtsCache});
+        break;
+      case 'xai':
+        audioData = await synthXai(logger, {
+          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
+          disableTtsCache});
+        break;
      case 'resemble':
        audioData = await synthResemble(logger, {
          credentials, stats, voice, key, text, options, renderForCaching, disableTtsStreaming, disableTtsCache});
@@ -316,6 +322,7 @@ const synthPolly = async(createHash, retrieveHash, logger,
    params += `,playback_id=${key}`;
    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
    params += ',vendor=aws';
+    params += `,voice=${voice}`;
    if (accessKeyId && secretAccessKey) {
      if (accessKeyId) params += `,accessKeyId=${accessKeyId}`;
      if (secretAccessKey) params += `,secretAccessKey=${secretAccessKey}`;
@@ -403,6 +410,32 @@ const synthPolly = async(createHash, retrieveHash, logger,
  }
 };

+/* google AudioConfig settings we support, as [google camelCase name, freeswitch param name] */
+const GOOGLE_AUDIO_SETTINGS = [
+  ['speakingRate', 'speaking_rate'],
+  ['pitch', 'pitch'],
+  ['volumeGainDb', 'volume_gain_db']
+];
+
+/**
+ * Extract google AudioConfig settings from the synthesizer options.  They may be supplied
+ * nested under an audioConfig property (mirroring google's AudioConfig object) or flat at
+ * the top level, and either google's camelCase or snake_case names are accepted.
+ * @see https://cloud.google.com/text-to-speech/docs/reference/rest/v1/text/synthesize#AudioConfig
+ * @returns object keyed by google's camelCase names, holding only valid numeric settings
+ */
+const googleAudioConfig = (options) => {
+  const provided = {...options, ...(options?.audioConfig || {})};
+  const audioConfig = {};
+  for (const [name, snakeName] of GOOGLE_AUDIO_SETTINGS) {
+    const value = provided[name] ?? provided[snakeName];
+    /* note: 0 is meaningful for pitch and volumeGainDb, so check for absence explicitly */
+    if (value === undefined || value === null || value === '') continue;
+    const num = Number(value);
+    if (Number.isFinite(num)) audioConfig[name] = num;
+  }
+  return audioConfig;
+};

 const synthGoogle = async(logger, {
  credentials, stats, language, voice, gender, key, text, model, options, instructions,
@@ -438,6 +471,15 @@ const synthGoogle = async(logger, {
    // comma is used to separate parameters in freeswitch tts module
    const prompt = options?.prompt || instructions;
    if (prompt) params += `,prompt=${prompt.replace(/\n/g, ' ').replace(/,/g, ';')}`;
+    /**
+     * AudioConfig settings.  Note google only honors these in some api modes:
+     * tts applies all of them, live (HD voices) applies speakingRate only,
+     * and gemini ignores them entirely (use prompt instead for style control).
+     */
+    const audioSettings = googleAudioConfig(options);
+    for (const [name, snakeName] of GOOGLE_AUDIO_SETTINGS) {
+      if (name in audioSettings) params += `,${snakeName}=${audioSettings[name]}`;
+    }
    params += '}';

    return {
@@ -506,6 +548,9 @@ const synthGoogle = async(logger, {
    sampleRate = 8000;
  }

+  /* gemini voices do not support the AudioConfig settings; they use prompt for style control */
+  if (!isGemini) Object.assign(audioConfig, googleAudioConfig(options));
+
  const opts = { input, voice: voiceParams, audioConfig };

  try {
@@ -724,70 +769,6 @@ const synthWellSaid = async(logger, {credentials, stats, language, voice, gender
  }
 };

-const synthNuance = async(client, logger, {credentials, stats, voice, model, text}) => {
-  let nuanceClient;
-  const {client_id, secret, nuance_tts_uri} = credentials;
-  if (nuance_tts_uri) {
-    nuanceClient = await createKryptonClient(nuance_tts_uri);
-  }
-  else {
-    /* get a nuance access token */
-    const {access_token} = await getNuanceAccessToken(client, logger, client_id, secret, 'tts');
-    nuanceClient = await createNuanceClient(access_token);
-  }
-
-  const v = new Voice();
-  const p = new AudioParameters();
-  const f = new AudioFormat();
-  const pcm = new PCM();
-  const params  = new EventParameters();
-  const request = new SynthesisRequest();
-  const input = new Input();
-
-  if (text.startsWith('<speak')) {
-    const ssml = new SSML();
-    ssml.setText(text);
-    input.setSsml(ssml);
-  }
-  else {
-    const t = new Text();
-    t.setText(text);
-    input.setText(t);
-  }
-  const sampleRate = 8000;
-  pcm.setSampleRateHz(sampleRate);
-  f.setPcm(pcm);
-  p.setAudioFormat(f);
-  v.setName(voice);
-  v.setModel(model);
-  request.setVoice(v);
-  request.setAudioParams(p);
-  request.setInput(input);
-  request.setEventParams(params);
-  request.setUserId('jambonz');
-
-  return new Promise((resolve, reject) => {
-    nuanceClient.unarySynthesize(request, (err, response) => {
-      if (err) {
-        console.error(err);
-        return reject(err);
-      }
-      const status = response.getStatus();
-      const code = status.getCode();
-      if (code !== 200) {
-        const message = status.getMessage();
-        const details = status.getDetails();
-        return reject({code, message, details});
-      }
-      resolve({
-        audioContent: Buffer.from(response.getAudio()),
-        extension: 'r8',
-        sampleRate
-      });
-    });
-  });
-};
-
 const synthNvidia = async(client, logger, {
  credentials, stats, language,  voice, model, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
 }) => {
@@ -953,101 +934,6 @@ const synthElevenlabs = async(logger, {
  }
 };

-const synthPlayHT = async(client, logger, {
-  credentials, options, stats, voice, language, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
-}) => {
-  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 = playht_tts_uri ? `${playht_tts_uri}/api/v2/tts/stream` : 'https://api.play.ht/api/v2/tts/stream';
-  // If model is play3.0, the synthesizeUrl is got from authentication endpoint
-  if (voice_engine === 'Play3.0') {
-    try {
-      const post = bent('https://api.play.ht', 'POST', 'json', 201, {
-        'AUTHORIZATION': api_key,
-        'X-USER-ID': user_id,
-        'Accept': 'application/json'
-      });
-      const key = makePlayhtKey(api_key);
-      const url = await client.get(key);
-      if (!url) {
-        const {inference_address, expires_at_ms} = await post('/api/v3/auth');
-        synthesizeUrl = inference_address;
-        const expiry =  Math.floor((expires_at_ms - Date.now()) / 1000 - 30);
-        await client.set(key, inference_address, 'EX', expiry);
-      } else {
-        // Use cached URL
-        synthesizeUrl = url;
-      }
-    } catch (err) {
-      logger.info({err}, 'synth PlayHT returned error for authentication version 3.0');
-      stats.increment('tts.count', ['vendor:playht', 'accepted:no']);
-      throw err;
-    }
-  }
-
-  /* default to using the streaming interface, unless disabled by env var OR we want just a cache file */
-  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
-    let params = '{';
-    params += `api_key=${api_key}`;
-    params += `,playback_id=${key}`;
-    params += `,user_id=${user_id}`;
-    params += ',vendor=playht';
-    params += `,voice=${voice}`;
-    params += `,voice_engine=${voice_engine}`;
-    params += `,synthesize_url=${synthesizeUrl}`;
-    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
-    params += `,language=${language}`;
-    if (opts.quality) params += `,quality=${opts.quality}`;
-    if (opts.speed) params += `,speed=${opts.speed}`;
-    if (opts.seed) params += `,style=${opts.seed}`;
-    if (opts.temperature) params += `,temperature=${opts.temperature}`;
-    if (opts.emotion) params += `,emotion=${opts.emotion}`;
-    if (opts.voice_guidance) params += `,voice_guidance=${opts.voice_guidance}`;
-    if (opts.style_guidance) params += `,style_guidance=${opts.style_guidance}`;
-    if (opts.text_guidance) params += `,text_guidance=${opts.text_guidance}`;
-    if (opts.top_p) params += `,top_p=${opts.top_p}`;
-    if (opts.repetition_penalty) params += `,repetition_penalty=${opts.repetition_penalty}`;
-    params += '}';
-
-    return {
-      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
-      servedFromCache: false,
-      rtt: 0
-    };
-  }
-
-  try {
-    const post = bent('POST', 'buffer', {
-      ...(voice_engine !== 'Play3.0' && {
-        'AUTHORIZATION': api_key,
-        'X-USER-ID': user_id,
-      }),
-      'Accept': 'audio/mpeg',
-      'Content-Type': 'application/json'
-    });
-
-    const audioContent = await post(synthesizeUrl, {
-      text,
-      ...(voice_engine === 'Play3.0' && { language }),
-      voice,
-      voice_engine,
-      output_format: 'mp3',
-      sample_rate: 8000,
-      ...opts
-    });
-    return {
-      audioContent,
-      extension: 'mp3',
-      sampleRate: 8000
-    };
-  } catch (err) {
-    logger.info({err}, 'synth PlayHT returned error');
-    stats.increment('tts.count', ['vendor:playht', 'accepted:no']);
-    throw err;
-  }
-};
-
 const synthInworld = async(logger, {
  credentials, options, stats, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
 }) => {
@@ -1064,8 +950,9 @@ const synthInworld = async(logger, {
    params += `,voice=${voice}`;
    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
    if (opts.temperature) params += `,temperature=${opts.temperature}`;
-    if (opts.audioConfig?.pitch) params += `,pitch=${opts.pitch}`;
-    if (opts.audioConfig?.speakingRate) params += `,speakingRate=${opts.speakingRate}`;
+    /* pitch and speakingRate are nested under audioConfig, matching Inworld's API */
+    if (opts.audioConfig?.pitch) params += `,pitch=${opts.audioConfig.pitch}`;
+    if (opts.audioConfig?.speakingRate) params += `,speakingRate=${opts.audioConfig.speakingRate}`;
    params += '}';

    return {
@@ -1173,54 +1060,72 @@ const synthRimelabs = async(logger, {
    throw err;
  }
 };
-const synthVerbio = async(client, logger, {
-  credentials, stats, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
+
+const synthMurf = async(logger, {
+  credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
 }) => {
-  //https://doc.speechcenter.verbio.com/#tag/Text-To-Speech-REST-API
-  if (text.length > 2000) {
-    throw new Error('Verbio cannot synthesize for the text length larger than 2000 characters');
-  }
-  const token = await getVerbioAccessToken(client, logger, credentials);
-  if (!process.env.JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+  const {api_key, model_id, api_uri, options: credOpts} = credentials;
+  const opts = !!options && Object.keys(options).length !== 0 ? options : JSON.parse(credOpts || '{}');
+
+  /* default to using the streaming interface, unless disabled by env var OR we want just a cache file */
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    /* param keys here must match mod_murf_tts's text_param handler */
    let params = '{';
-    params += `access_token=${token.access_token}`;
+    params += `api_key=${api_key}`;
    params += `,playback_id=${key}`;
-    params += ',vendor=verbio';
+    params += ',vendor=murf';
    params += `,voice=${voice}`;
+    if (model_id) params += `,model_id=${model_id}`;
+    if (language) params += `,language=${language}`;
+    if (api_uri) params += `,api_uri=${api_uri}`;
+    if (opts.style) params += `,style=${opts.style}`;
+    if (opts.rate !== undefined && opts.rate !== null) params += `,rate=${opts.rate}`;
+    if (opts.pitch !== undefined && opts.pitch !== null) params += `,pitch=${opts.pitch}`;
+    if (opts.variation !== undefined && opts.variation !== null) params += `,variation=${opts.variation}`;
    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
    params += '}';

    return {
-      filePath: `say:${params}${text.replace(/\n/g, ' ')}`,
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
      servedFromCache: false,
      rtt: 0
    };
  }

  try {
-    const post = bent('https://us.rest.speechcenter.verbio.com', 'POST', 'buffer', {
-      'Authorization': `Bearer ${token.access_token}`,
-      'User-Agent': 'jambonz',
+    const sampleRate = 8000;
+    /* no Accept header: murf returns 406 if it doesn't match; the response
+       container is selected by the `format` field in the body instead */
+    const post = bent(api_uri || 'https://global.api.murf.ai', 'POST', 'buffer', {
+      'api-key': api_key,
      'Content-Type': 'application/json'
    });
-    const audioContent = await post('/api/v1/synthesize', {
+    /* murf REST schema is documented loosely; field names follow the SDK params
+       (voice_id/model/format/sample_rate) plus the websocket voice fields. */
+    const audioContent = await post('/v1/speech/stream', {
+      text,
      voice_id: voice,
-      output_sample_rate: '8k',
-      output_encoding: 'pcm16',
-      text
+      ...(model_id && {model: model_id}),
+      ...(language && {locale: language}),
+      ...(opts.style && {style: opts.style}),
+      ...(opts.rate !== undefined && opts.rate !== null && {rate: opts.rate}),
+      ...(opts.pitch !== undefined && opts.pitch !== null && {pitch: opts.pitch}),
+      ...(opts.variation !== undefined && opts.variation !== null && {variation: opts.variation}),
+      format: 'WAV',
+      sample_rate: sampleRate,
+      channel_type: 'MONO'
    });
    return {
      audioContent,
-      extension: 'r8',
-      sampleRate: 8000
+      extension: 'wav',
+      sampleRate
    };
  } catch (err) {
-    logger.info({err}, 'synth Verbio returned error');
-    stats.increment('tts.count', ['vendor:verbio', 'accepted:no']);
+    logger.info({err}, 'synth murf returned error');
+    stats.increment('tts.count', ['vendor:murf', 'accepted:no']);
    throw err;
  }
 };
-
 const synthWhisper = async(logger, {credentials, stats, voice, key, text, instructions,
  renderForCaching, disableTtsStreaming, disableTtsCache}) => {
  const {api_key, model_id, baseURL, timeout, speed} = credentials;
@@ -1311,6 +1216,116 @@ const synthDeepgram = async(logger, {credentials, stats, model, key, text, rende
  }
 };

+// Deepgram Flux TTS — the conversation-native model served from /v2/speak.
+// Streaming rides the mediajam deepgramflux dialect via a say: filePath; the
+// batch/cache path POSTs to /v2/speak (mp3 is batch-only for Flux).
+const synthDeepgramFlux = async(logger, {credentials, stats, model, key, text, renderForCaching,
+  disableTtsStreaming, disableTtsCache}) => {
+  const {api_key, deepgram_tts_uri} = credentials;
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    let params = '{';
+    params += `api_key=${api_key}`;
+    params += `,playback_id=${key}`;
+    params += ',vendor=deepgramflux';
+    params += `,voice=${model}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    if (deepgram_tts_uri) params += `,endpoint=${deepgram_tts_uri}`;
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+  try {
+    const post = bent(deepgram_tts_uri || 'https://api.deepgram.com', 'POST', 'buffer', {
+      // on-premise deepgram does not require to have api_key
+      ...(api_key && {'Authorization': `Token ${api_key}`}),
+      'Accept': 'audio/mpeg',
+      'Content-Type': 'application/json'
+    });
+    const audioContent = await post(`/v2/speak?model=${model}`, {
+      text
+    });
+    return {
+      audioContent,
+      extension: 'mp3',
+      sampleRate: 8000
+    };
+  } catch (err) {
+    logger.info({err}, 'synth Deepgram Flux returned error');
+    stats.increment('tts.count', ['vendor:deepgramflux', 'accepted:no']);
+    throw err;
+  }
+};
+
+const synthXai = async(logger, {
+  credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
+}) => {
+  const {api_key, api_uri, options: credOpts} = credentials;
+  const opts = !!options && Object.keys(options).length !== 0 ? options : JSON.parse(credOpts || '{}');
+  const speed = opts.speed;
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    let params = '{';
+    params += `api_key=${api_key}`;
+    params += `,playback_id=${key}`;
+    params += ',vendor=xai';
+    if (voice) params += `,voice=${voice}`;
+    if (language) params += `,language=${language}`;
+    if (speed !== null && speed !== undefined) params += `,speed=${speed}`;
+    if (opts.optimize_streaming_latency != null) {
+      params += `,optimize_streaming_latency=${opts.optimize_streaming_latency}`;
+    }
+    if (opts.text_normalization != null) params += `,text_normalization=${opts.text_normalization}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    if (api_uri) params += `,endpoint=${api_uri}`;
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+  try {
+    const post = bent(`https://${api_uri || 'api.x.ai'}`, 'POST', 'buffer', {
+      'Authorization': `Bearer ${api_key}`,
+      'Content-Type': 'application/json'
+    });
+    const audioContent = await post('/v1/tts', {
+      text,
+      language: language || 'auto',
+      ...(voice && {voice_id: voice}),
+      ...(speed !== null && speed !== undefined && {speed}),
+      ...(opts.optimize_streaming_latency != null && {optimize_streaming_latency: opts.optimize_streaming_latency}),
+      ...(opts.text_normalization != null && {text_normalization: opts.text_normalization}),
+      output_format: {
+        codec: 'wav',
+        sample_rate: 8000
+      }
+    });
+    return {
+      audioContent,
+      extension: 'wav',
+      sampleRate: 8000
+    };
+  } catch (err) {
+    // xAI errors are JSON {code, error} - read the body so the surfaced message isn't 'undefined'
+    if (err.name === 'StatusError' && typeof err.text === 'function') {
+      try {
+        const body = await err.text();
+        if (body) err.message = body;
+      } catch (readErr) {
+        logger.info({readErr}, 'synth xai: failed to read error response body');
+      }
+    }
+    logger.info({err}, 'synth xai returned error');
+    stats.increment('tts.count', ['vendor:xai', 'accepted:no']);
+    throw err;
+  }
+};
+
 const synthCartesia = async(logger, {
  credentials, options, stats, voice, language, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
 }) => {
@@ -1436,6 +1451,197 @@ const synthCartesia = async(logger, {

 };

+/* gradium.ai — json websocket for streaming, and a POST endpoint for the cache
+   render. only_audio:true + pcm_8000 returns bare little-endian 16-bit samples,
+   which is exactly the r8 container, so we avoid gradium's streaming wav header
+   (it carries 0xffffffff as the RIFF size, since length is unknown up front).
+*/
+const synthGradium = async(logger, {
+  credentials, options, stats, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
+}) => {
+  const {api_key, model_id} = credentials;
+  const {json_config, pronunciation_id} = options || {};
+
+  /* default to using the streaming interface, unless disabled by env var OR we want just a cache file */
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    let params = '{';
+    params += `api_key=${api_key}`;
+    params += `,playback_id=${key}`;
+    params += ',vendor=gradium';
+    params += `,voice=${voice}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    if (model_id) params += `,model_id=${model_id}`;
+    if (pronunciation_id) params += `,pronunciation_id=${pronunciation_id}`;
+    /* the say: param parser is brace-aware, so a nested json object survives intact */
+    if (json_config) {
+      params += `,json_config=${typeof json_config === 'string' ? json_config : JSON.stringify(json_config)}`;
+    }
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+
+  try {
+    const sampleRate = 8000;
+    const post = bent('https://api.gradium.ai', 'POST', 'buffer', {
+      'x-api-key': api_key,
+      'Content-Type': 'application/json'
+    });
+    const audioContent = await post('/api/post/speech/tts', {
+      text,
+      voice_id: voice,
+      model_name: model_id || 'default',
+      output_format: `pcm_${sampleRate}`,
+      only_audio: true,
+      ...(json_config && {json_config}),
+      ...(pronunciation_id && {pronunciation_id})
+    });
+    return {
+      audioContent,
+      extension: 'r8',
+      sampleRate
+    };
+  } catch (err) {
+    logger.info({err}, 'synth gradium returned error');
+    stats.increment('tts.count', ['vendor:gradium', 'accepted:no']);
+    throw err;
+  }
+};
+
+/* nineninesix.ai — a Cartesia-compatible API, but only raw/wav come back
+   (mp3 is rejected), so the cache render asks for wav rather than mp3. */
+/* fish.audio — msgpack websocket for streaming, and a POST endpoint for the cache
+   render. format:pcm + sample_rate:8000 returns bare little-endian 16-bit samples,
+   which is exactly the r8 container. we avoid fish's wav output because its RIFF
+   header carries a placeholder size (0xffffff24) — length is unknown up front, as
+   with gradium.
+
+   fish is a voice-cloning vendor: the "voice" is a reference_id returned by
+   POST /model, and omitting it entirely synthesizes with fish's default voice.
+   the sentinel value 'default' (the bundled fallback entry in the portal) means
+   exactly that — send no reference_id.
+*/
+const synthFishaudio = async(logger, {
+  credentials, options, stats, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
+}) => {
+  const {api_key, model_id, fishaudio_tts_uri} = credentials;
+  const {reference_id, latency, chunk_length, speed, volume} = options || {};
+
+  /* free-text reference_id in the vendor options wins over the voice selector */
+  const refId = reference_id || (voice && voice !== 'default' ? voice : null);
+
+  /* default to using the streaming interface, unless disabled by env var OR we want just a cache file */
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    let params = '{';
+    params += `api_key=${api_key}`;
+    params += `,playback_id=${key}`;
+    params += ',vendor=fishaudio';
+    params += `,voice=${refId || 'default'}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    if (model_id) params += `,model_id=${model_id}`;
+    if (latency) params += `,latency=${latency}`;
+    if (chunk_length) params += `,chunk_length=${chunk_length}`;
+    if (speed) params += `,speed=${speed}`;
+    if (volume) params += `,volume=${volume}`;
+    if (fishaudio_tts_uri) params += `,endpoint=${fishaudio_tts_uri}`;
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+
+  try {
+    const sampleRate = 8000;
+    const post = bent(fishaudio_tts_uri || 'https://api.fish.audio', 'POST', 'buffer', {
+      'Authorization': `Bearer ${api_key}`,
+      'Content-Type': 'application/json',
+      /* the model is selected by header, not in the body */
+      'model': model_id || 's2.1-pro'
+    });
+    const audioContent = await post('/v1/tts', {
+      text,
+      format: 'pcm',
+      sample_rate: sampleRate,
+      ...(refId && {reference_id: refId}),
+      ...(latency && {latency}),
+      ...(chunk_length && {chunk_length: parseInt(chunk_length, 10)}),
+      ...((speed || volume) && {prosody: {
+        ...(speed && {speed: parseFloat(speed)}),
+        ...(volume && {volume: parseFloat(volume)})
+      }})
+    });
+    return {
+      audioContent,
+      extension: 'r8',
+      sampleRate
+    };
+  } catch (err) {
+    logger.info({err}, 'synth fishaudio returned error');
+    stats.increment('tts.count', ['vendor:fishaudio', 'accepted:no']);
+    throw err;
+  }
+};
+
+const synthNineninesix = async(logger, {
+  credentials, stats, voice, language, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
+}) => {
+  const {api_key, model_id} = credentials;
+
+  /* default to using the streaming interface, unless disabled by env var OR we want just a cache file */
+  if (!JAMBONES_DISABLE_TTS_STREAMING && !renderForCaching && !disableTtsStreaming) {
+    let params = '{';
+    params += `api_key=${api_key}`;
+    params += `,playback_id=${key}`;
+    params += `,model_id=${model_id}`;
+    params += ',vendor=nineninesix';
+    params += `,voice=${voice}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    if (language) params += `,language=${language}`;
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+
+  try {
+    const sampleRate = 8000;
+    const post = bent('https://api.nineninesix.ai', 'POST', 'buffer', {
+      'Authorization': `Bearer ${api_key}`,
+      'Content-Type': 'application/json'
+    });
+    const audioContent = await post('/tts/bytes', {
+      model_id,
+      transcript: text,
+      voice: {mode: 'id', id: voice},
+      ...(language && {language}),
+      output_format: {
+        container: 'wav',
+        encoding: 'pcm_s16le',
+        sample_rate: sampleRate
+      }
+    });
+    return {
+      audioContent,
+      extension: 'wav',
+      sampleRate
+    };
+  } catch (err) {
+    logger.info({err}, 'synth nineninesix returned error');
+    stats.increment('tts.count', ['vendor:nineninesix', 'accepted:no']);
+    throw err;
+  }
+};
+
 const synthResemble = async(logger, {
  credentials, options, stats, voice, key, text, renderForCaching, disableTtsStreaming, disableTtsCache
 }) => {
@@ -1,20 +1,7 @@
 const crypto = require('crypto');
-const {SynthesizerClient} = require('../stubs/nuance/synthesizer_grpc_pb');
 const {RivaSpeechSynthesisClient} = require('../stubs/riva/proto/riva_tts_grpc_pb');
-const {Pool} = require('undici');
-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 { TMP_FOLDER, HTTP_TIMEOUT } = require('./config');
-
-const debug = require('debug')('jambonz:realtimedb-helpers');
-/**
- * Future TODO: cache recently used connections to providers
- * to avoid connection overhead during a call.
- * Will need to periodically age them out to avoid memory leaks.
- */
-//const nuanceClientMap = new Map();
+const { TMP_FOLDER } = require('./config');

 function makeSynthKey({
  account_sid = '',
@@ -45,90 +32,12 @@ const noopLogger = {
  error: () => {}
 };

-const toBase64 = (str) => Buffer.from(str || '', 'utf8').toString('base64');
-
-function makeBasicAuthHeader(username, password) {
-  if (!username || !password) return {};
-  const creds = `${encodeURIComponent(username)}:${password || ''}`;
-  const header = `Basic ${toBase64(creds)}`;
-  return {Authorization: header};
-}
-
 function makeAwsKey(awsAccessKeyId) {
  const hash = crypto.createHash('sha1');
  hash.update(awsAccessKeyId);
  return `aws:${hash.digest('hex')}`;
 }

-function makePlayhtKey(apiKey) {
-  const hash = crypto.createHash('sha1');
-  hash.update(apiKey);
-  return `playht:${hash.digest('hex')}`;
-}
-function makeVerbioKey(client_id) {
-  const hash = crypto.createHash('sha1');
-  hash.update(client_id);
-  return `verbio:${hash.digest('hex')}`;
-}
-
-function makeNuanceKey(clientId, secret, scope) {
-  const hash = crypto.createHash('sha1');
-  hash.update(`${clientId}:${secret}:${scope}`);
-  return `nuance:${hash.digest('hex')}`;
-}
-
-const getNuanceAccessToken = async(clientId, secret, scope = 'asr tts') => {
-  const payload = {
-    grant_type: 'client_credentials',
-    scope
-  };
-  const auth = makeBasicAuthHeader(clientId, secret);
-  const {statusCode, headers, body} =  await pool.request({
-    path: '/oauth2/token',
-    method: 'POST',
-    headers: {
-      ...auth,
-      'Content-Type': 'application/x-www-form-urlencoded'
-    },
-    body: formurlencoded(payload),
-    timeout: HTTP_TIMEOUT,
-    followRedirects: false
-  });
-
-  if (200 !== statusCode) {
-    debug({statusCode, headers, body: body.text()}, 'error fetching access token from Nuance');
-    const err = new Error();
-    err.statusCode = statusCode;
-    throw err;
-  }
-  const json = await body.json();
-  return json.access_token;
-};
-
-const createKryptonClient = async(uri) => {
-  const client = new SynthesizerClient(uri, grpc.credentials.createInsecure());
-  return client;
-};
-
-const createNuanceClient = async(access_token) => {
-
-  //if (nuanceClientMap.has(access_token)) return nuanceClientMap.get(access_token);
-
-  const generateMetadata = (params, callback) => {
-    var metadata = new grpc.Metadata();
-    metadata.add('authorization', `Bearer ${access_token}`);
-    callback(null, metadata);
-  };
-
-  const sslCreds = grpc.credentials.createSsl();
-  const authCreds = grpc.credentials.createFromMetadataGenerator(generateMetadata);
-  const combined_creds = grpc.credentials.combineChannelCredentials(sslCreds, authCreds);
-  const client = new SynthesizerClient(NUANCE_AUTH_ENDPOINT, combined_creds);
-
-  //if (process.env.NUANCE_CACHE_TTS_CONNECTIONS) nuanceClientMap.set(access_token, client);
-  return client;
-};
-
 const createRivaClient = async(rivaUri) => {
  const client = new RivaSpeechSynthesisClient(rivaUri, grpc.credentials.createInsecure());
  return client;
@@ -136,17 +45,8 @@ const createRivaClient = async(rivaUri) => {

 module.exports = {
  makeSynthKey,
-  makeNuanceKey,
-
-  makePlayhtKey,
  makeAwsKey,
-  makeVerbioKey,
-  getNuanceAccessToken,
-  createNuanceClient,
-  createKryptonClient,
  createRivaClient,
-  makeBasicAuthHeader,
-  NUANCE_AUTH_ENDPOINT,
  noopLogger,
  makeFilePath
 };
@@ -1,15 +1,14 @@
 {
  "name": "@jambonz/speech-utils",
-  "version": "1.0.4",
+  "version": "1.0.18",
  "lockfileVersion": 2,
  "requires": true,
  "packages": {
    "": {
      "name": "@jambonz/speech-utils",
-      "version": "1.0.4",
+      "version": "1.0.18",
      "license": "MIT",
      "dependencies": {
-        "23": "^0.0.0",
        "@aws-sdk/client-polly": "^3.496.0",
        "@aws-sdk/client-sts": "^3.496.0",
        "@cartesia/cartesia-js": "^2.2.7",
@@ -18,11 +17,9 @@
        "@jambonz/realtimedb-helpers": "^0.8.7",
        "bent": "^7.3.12",
        "debug": "^4.3.4",
-        "form-urlencoded": "^6.1.4",
        "google-protobuf": "^3.21.2",
-        "microsoft-cognitiveservices-speech-sdk": "1.38.0",
-        "openai": "^4.98.0",
-        "undici": "^7.5.0"
+        "microsoft-cognitiveservices-speech-sdk": "^1.51.0",
+        "openai": "^4.98.0"
      },
      "devDependencies": {
        "config": "^4.2.0",
@@ -761,6 +758,46 @@
        "node": ">=18.0.0"
      }
    },
+    "node_modules/@azure/abort-controller": {
+      "version": "2.2.0",
+      "resolved": "https://registry.npmjs.org/@azure/abort-controller/-/abort-controller-2.2.0.tgz",
+      "integrity": "sha512-fNAjWnA/nZ2jz31kxR/AqRaUT8ewHBw/WuBIosK0moMy1C9e5ValbDfFdIxJzVOOYaYkV/b2F1S4H/aHiqfVQg==",
+      "license": "MIT",
+      "dependencies": {
+        "tslib": "^2.6.2"
+      },
+      "engines": {
+        "node": ">=22.0.0"
+      }
+    },
+    "node_modules/@azure/core-auth": {
+      "version": "1.11.0",
+      "resolved": "https://registry.npmjs.org/@azure/core-auth/-/core-auth-1.11.0.tgz",
+      "integrity": "sha512-IUZydyTUkDnYdstOW9pFOOUQlBjAepK5teihDE3x6yxsPJs/hsAaaYpeGxdxrgtOiJbBKSjKW7MDk7AEhb4LRg==",
+      "license": "MIT",
+      "dependencies": {
+        "@azure/abort-controller": "^2.1.2",
+        "@azure/core-util": "^1.13.0",
+        "tslib": "^2.6.2"
+      },
+      "engines": {
+        "node": ">=22.0.0"
+      }
+    },
+    "node_modules/@azure/core-util": {
+      "version": "1.14.0",
+      "resolved": "https://registry.npmjs.org/@azure/core-util/-/core-util-1.14.0.tgz",
+      "integrity": "sha512-9n2pWK61veAuN0V20t9lOuoV4CFMdyAZ1ygZzvBGk/pBBJRib/PjL9PLXa/aI2CcPpyHfqVsxxqLCYl6uZlfDw==",
+      "license": "MIT",
+      "dependencies": {
+        "@azure/abort-controller": "^2.1.2",
+        "@typespec/ts-http-runtime": "^0.3.0",
+        "tslib": "^2.6.2"
+      },
+      "engines": {
+        "node": ">=22.0.0"
+      }
+    },
    "node_modules/@babel/code-frame": {
      "version": "7.26.2",
      "resolved": "https://registry.npmjs.org/@babel/code-frame/-/code-frame-7.26.2.tgz",
@@ -2392,11 +2429,19 @@
      "resolved": "https://registry.npmjs.org/@types/webrtc/-/webrtc-0.0.37.tgz",
      "integrity": "sha512-JGAJC/ZZDhcrrmepU4sPLQLIOIAgs5oIK+Ieq90K8fdaNMhfdfqmYatJdgif1NDQtvrSlTOGJDUYHIDunuufOg=="
    },
-    "node_modules/23": {
-      "version": "0.0.0",
-      "resolved": "https://registry.npmjs.org/23/-/23-0.0.0.tgz",
-      "integrity": "sha512-uAETf9Okr72trtp1pNXYKhFCTTI1EKGcYMA8gw3jLGhlbaDX+grrNToEWrpt8luxRAvrZWBTvyB5wk3PpNjGQQ==",
-      "license": "ISC"
+    "node_modules/@typespec/ts-http-runtime": {
+      "version": "0.3.8",
+      "resolved": "https://registry.npmjs.org/@typespec/ts-http-runtime/-/ts-http-runtime-0.3.8.tgz",
+      "integrity": "sha512-bLMpVcWZNzq6lYOybwFwOAR1IXKcHnhUNqYeHjl1bET/qE3jFPFH+p8Wrh3rU4xwdnifPxmKNESBYnvnmc75aA==",
+      "license": "MIT",
+      "dependencies": {
+        "http-proxy-agent": "^7.0.0",
+        "https-proxy-agent": "^7.0.0",
+        "tslib": "^2.6.2"
+      },
+      "engines": {
+        "node": ">=22.0.0"
+      }
    },
    "node_modules/abort-controller": {
      "version": "3.0.0",
@@ -3747,11 +3792,6 @@
      "resolved": "https://registry.npmjs.org/form-data-encoder/-/form-data-encoder-1.7.2.tgz",
      "integrity": "sha512-qfqtYan3rxrnCk1VYaA4H+Ms9xdpPqvLZa6xmMgFvhO32x7/3J/ExcTd6qpxM0vH2GdMI+poehyBZvqfMTto8A=="
    },
-    "node_modules/form-urlencoded": {
-      "version": "6.1.4",
-      "resolved": "https://registry.npmjs.org/form-urlencoded/-/form-urlencoded-6.1.4.tgz",
-      "integrity": "sha512-vDFWgogZ04th20oxQtDSF8CDjrBgAPtP1V+W3kZni5IZqyasDDVNqCQGOXfYlmOJj0zpROgiAtPJXCmidtxDTQ=="
-    },
    "node_modules/formdata-node": {
      "version": "4.4.1",
      "resolved": "https://registry.npmjs.org/formdata-node/-/formdata-node-4.4.1.tgz",
@@ -5284,17 +5324,18 @@
      }
    },
    "node_modules/microsoft-cognitiveservices-speech-sdk": {
-      "version": "1.38.0",
-      "resolved": "https://registry.npmjs.org/microsoft-cognitiveservices-speech-sdk/-/microsoft-cognitiveservices-speech-sdk-1.38.0.tgz",
-      "integrity": "sha512-NA6J4eIDkeR9iN83rcn77Kn5AWQcizDEn1tLMjzRvSovUNB1FrZe0mWYO0fsGltUwMl3Ns5OZ3lGw42PU4fEYA==",
+      "version": "1.51.0",
+      "resolved": "https://registry.npmjs.org/microsoft-cognitiveservices-speech-sdk/-/microsoft-cognitiveservices-speech-sdk-1.51.0.tgz",
+      "integrity": "sha512-BLLovv5PegOr5Lp52h4CSgL2c/ViFAh9LoU49ILNPyb/Z0giGI7nPV3xNLcmTtHLT+kfYNRGUIA62vORLzP8TA==",
      "license": "MIT",
      "dependencies": {
+        "@azure/core-auth": "^1.9.0",
        "@types/webrtc": "^0.0.37",
        "agent-base": "^6.0.1",
        "bent": "^7.3.12",
        "https-proxy-agent": "^4.0.0",
-        "uuid": "^9.0.0",
-        "ws": "^7.5.6"
+        "uuid": "^11.1.1",
+        "ws": "^8.21.0"
      }
    },
    "node_modules/microsoft-cognitiveservices-speech-sdk/node_modules/https-proxy-agent": {
@@ -5318,36 +5359,16 @@
      }
    },
    "node_modules/microsoft-cognitiveservices-speech-sdk/node_modules/uuid": {
-      "version": "9.0.1",
-      "resolved": "https://registry.npmjs.org/uuid/-/uuid-9.0.1.tgz",
-      "integrity": "sha512-b+1eJOlsR9K8HJpow9Ok3fiWOWSIcIzXodvv0rQjVoOVNpWMpxf1wZNpt4y9h10odCNrqnYp1OBzRktckBe3sA==",
+      "version": "11.1.1",
+      "resolved": "https://registry.npmjs.org/uuid/-/uuid-11.1.1.tgz",
+      "integrity": "sha512-vIYxrBCC/N/K+Js3qSN88go7kIfNPssr/hHCesKCQNAjmgvYS2oqr69kIufEG+O4+PfezOH4EbIeHCfFov8ZgQ==",
      "funding": [
        "https://github.com/sponsors/broofa",
        "https://github.com/sponsors/ctavan"
      ],
-      "bin": {
-        "uuid": "dist/bin/uuid"
-      }
-    },
-    "node_modules/microsoft-cognitiveservices-speech-sdk/node_modules/ws": {
-      "version": "7.5.10",
-      "resolved": "https://registry.npmjs.org/ws/-/ws-7.5.10.tgz",
-      "integrity": "sha512-+dbF1tHwZpXcbOJdVOkzLDxZP1ailvSxM6ZweXTegylPny803bFhA+vqBYw4s31NSAk4S2Qz+AKXK9a4wkdjcQ==",
      "license": "MIT",
-      "engines": {
-        "node": ">=8.3.0"
-      },
-      "peerDependencies": {
-        "bufferutil": "^4.0.1",
-        "utf-8-validate": "^5.0.2"
-      },
-      "peerDependenciesMeta": {
-        "bufferutil": {
-          "optional": true
-        },
-        "utf-8-validate": {
-          "optional": true
-        }
+      "bin": {
+        "uuid": "dist/esm/bin/uuid"
      }
    },
    "node_modules/mime-db": {
@@ -7037,15 +7058,6 @@
        "url": "https://github.com/sponsors/ljharb"
      }
    },
-    "node_modules/undici": {
-      "version": "7.24.5",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-7.24.5.tgz",
-      "integrity": "sha512-3IWdCpjgxp15CbJnsi/Y9TCDE7HWVN19j1hmzVhoAkY/+CJx449tVxT5wZc1Gwg8J+P0LWvzlBzxYRnHJ+1i7Q==",
-      "license": "MIT",
-      "engines": {
-        "node": ">=20.18.1"
-      }
-    },
    "node_modules/undici-types": {
      "version": "5.26.5",
      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-5.26.5.tgz",
@@ -7348,11 +7360,6 @@
    }
  },
  "dependencies": {
-    "23": {
-      "version": "0.0.0",
-      "resolved": "https://registry.npmjs.org/23/-/23-0.0.0.tgz",
-      "integrity": "sha512-uAETf9Okr72trtp1pNXYKhFCTTI1EKGcYMA8gw3jLGhlbaDX+grrNToEWrpt8luxRAvrZWBTvyB5wk3PpNjGQQ=="
-    },
    "@aashutoshrathi/word-wrap": {
      "version": "1.2.6",
      "resolved": "https://registry.npmjs.org/@aashutoshrathi/word-wrap/-/word-wrap-1.2.6.tgz",
@@ -7930,6 +7937,34 @@
      "resolved": "https://registry.npmjs.org/@aws/lambda-invoke-store/-/lambda-invoke-store-0.2.4.tgz",
      "integrity": "sha512-iY8yvjE0y651BixKNPgmv1WrQc+GZ142sb0z4gYnChDDY2YqI4P/jsSopBWrKfAt7LOJAkOXt7rC/hms+WclQQ=="
    },
+    "@azure/abort-controller": {
+      "version": "2.2.0",
+      "resolved": "https://registry.npmjs.org/@azure/abort-controller/-/abort-controller-2.2.0.tgz",
+      "integrity": "sha512-fNAjWnA/nZ2jz31kxR/AqRaUT8ewHBw/WuBIosK0moMy1C9e5ValbDfFdIxJzVOOYaYkV/b2F1S4H/aHiqfVQg==",
+      "requires": {
+        "tslib": "^2.6.2"
+      }
+    },
+    "@azure/core-auth": {
+      "version": "1.11.0",
+      "resolved": "https://registry.npmjs.org/@azure/core-auth/-/core-auth-1.11.0.tgz",
+      "integrity": "sha512-IUZydyTUkDnYdstOW9pFOOUQlBjAepK5teihDE3x6yxsPJs/hsAaaYpeGxdxrgtOiJbBKSjKW7MDk7AEhb4LRg==",
+      "requires": {
+        "@azure/abort-controller": "^2.1.2",
+        "@azure/core-util": "^1.13.0",
+        "tslib": "^2.6.2"
+      }
+    },
+    "@azure/core-util": {
+      "version": "1.14.0",
+      "resolved": "https://registry.npmjs.org/@azure/core-util/-/core-util-1.14.0.tgz",
+      "integrity": "sha512-9n2pWK61veAuN0V20t9lOuoV4CFMdyAZ1ygZzvBGk/pBBJRib/PjL9PLXa/aI2CcPpyHfqVsxxqLCYl6uZlfDw==",
+      "requires": {
+        "@azure/abort-controller": "^2.1.2",
+        "@typespec/ts-http-runtime": "^0.3.0",
+        "tslib": "^2.6.2"
+      }
+    },
    "@babel/code-frame": {
      "version": "7.26.2",
      "resolved": "https://registry.npmjs.org/@babel/code-frame/-/code-frame-7.26.2.tgz",
@@ -9124,6 +9159,16 @@
      "resolved": "https://registry.npmjs.org/@types/webrtc/-/webrtc-0.0.37.tgz",
      "integrity": "sha512-JGAJC/ZZDhcrrmepU4sPLQLIOIAgs5oIK+Ieq90K8fdaNMhfdfqmYatJdgif1NDQtvrSlTOGJDUYHIDunuufOg=="
    },
+    "@typespec/ts-http-runtime": {
+      "version": "0.3.8",
+      "resolved": "https://registry.npmjs.org/@typespec/ts-http-runtime/-/ts-http-runtime-0.3.8.tgz",
+      "integrity": "sha512-bLMpVcWZNzq6lYOybwFwOAR1IXKcHnhUNqYeHjl1bET/qE3jFPFH+p8Wrh3rU4xwdnifPxmKNESBYnvnmc75aA==",
+      "requires": {
+        "http-proxy-agent": "^7.0.0",
+        "https-proxy-agent": "^7.0.0",
+        "tslib": "^2.6.2"
+      }
+    },
    "abort-controller": {
      "version": "3.0.0",
      "resolved": "https://registry.npmjs.org/abort-controller/-/abort-controller-3.0.0.tgz",
@@ -10064,11 +10109,6 @@
      "resolved": "https://registry.npmjs.org/form-data-encoder/-/form-data-encoder-1.7.2.tgz",
      "integrity": "sha512-qfqtYan3rxrnCk1VYaA4H+Ms9xdpPqvLZa6xmMgFvhO32x7/3J/ExcTd6qpxM0vH2GdMI+poehyBZvqfMTto8A=="
    },
-    "form-urlencoded": {
-      "version": "6.1.4",
-      "resolved": "https://registry.npmjs.org/form-urlencoded/-/form-urlencoded-6.1.4.tgz",
-      "integrity": "sha512-vDFWgogZ04th20oxQtDSF8CDjrBgAPtP1V+W3kZni5IZqyasDDVNqCQGOXfYlmOJj0zpROgiAtPJXCmidtxDTQ=="
-    },
    "formdata-node": {
      "version": "4.4.1",
      "resolved": "https://registry.npmjs.org/formdata-node/-/formdata-node-4.4.1.tgz",
@@ -11117,16 +11157,17 @@
      "integrity": "sha512-/IXtbwEk5HTPyEwyKX6hGkYXxM9nbj64B+ilVJnC/R6B0pH5G4V3b0pVbL7DBj4tkhBAppbQUlf6F6Xl9LHu1g=="
    },
    "microsoft-cognitiveservices-speech-sdk": {
-      "version": "1.38.0",
-      "resolved": "https://registry.npmjs.org/microsoft-cognitiveservices-speech-sdk/-/microsoft-cognitiveservices-speech-sdk-1.38.0.tgz",
-      "integrity": "sha512-NA6J4eIDkeR9iN83rcn77Kn5AWQcizDEn1tLMjzRvSovUNB1FrZe0mWYO0fsGltUwMl3Ns5OZ3lGw42PU4fEYA==",
+      "version": "1.51.0",
+      "resolved": "https://registry.npmjs.org/microsoft-cognitiveservices-speech-sdk/-/microsoft-cognitiveservices-speech-sdk-1.51.0.tgz",
+      "integrity": "sha512-BLLovv5PegOr5Lp52h4CSgL2c/ViFAh9LoU49ILNPyb/Z0giGI7nPV3xNLcmTtHLT+kfYNRGUIA62vORLzP8TA==",
      "requires": {
+        "@azure/core-auth": "^1.9.0",
        "@types/webrtc": "^0.0.37",
        "agent-base": "^6.0.1",
        "bent": "^7.3.12",
        "https-proxy-agent": "^4.0.0",
-        "uuid": "^9.0.0",
-        "ws": "^7.5.6"
+        "uuid": "^11.1.1",
+        "ws": "^8.21.0"
      },
      "dependencies": {
        "https-proxy-agent": {
@@ -11146,15 +11187,9 @@
          }
        },
        "uuid": {
-          "version": "9.0.1",
-          "resolved": "https://registry.npmjs.org/uuid/-/uuid-9.0.1.tgz",
-          "integrity": "sha512-b+1eJOlsR9K8HJpow9Ok3fiWOWSIcIzXodvv0rQjVoOVNpWMpxf1wZNpt4y9h10odCNrqnYp1OBzRktckBe3sA=="
-        },
-        "ws": {
-          "version": "7.5.10",
-          "resolved": "https://registry.npmjs.org/ws/-/ws-7.5.10.tgz",
-          "integrity": "sha512-+dbF1tHwZpXcbOJdVOkzLDxZP1ailvSxM6ZweXTegylPny803bFhA+vqBYw4s31NSAk4S2Qz+AKXK9a4wkdjcQ==",
-          "requires": {}
+          "version": "11.1.1",
+          "resolved": "https://registry.npmjs.org/uuid/-/uuid-11.1.1.tgz",
+          "integrity": "sha512-vIYxrBCC/N/K+Js3qSN88go7kIfNPssr/hHCesKCQNAjmgvYS2oqr69kIufEG+O4+PfezOH4EbIeHCfFov8ZgQ=="
        }
      }
    },
@@ -12358,11 +12393,6 @@
        "which-boxed-primitive": "^1.0.2"
      }
    },
-    "undici": {
-      "version": "7.24.5",
-      "resolved": "https://registry.npmjs.org/undici/-/undici-7.24.5.tgz",
-      "integrity": "sha512-3IWdCpjgxp15CbJnsi/Y9TCDE7HWVN19j1hmzVhoAkY/+CJx449tVxT5wZc1Gwg8J+P0LWvzlBzxYRnHJ+1i7Q=="
-    },
    "undici-types": {
      "version": "5.26.5",
      "resolved": "https://registry.npmjs.org/undici-types/-/undici-types-5.26.5.tgz",
@@ -1,6 +1,6 @@
 {
  "name": "@jambonz/speech-utils",
-  "version": "1.0.4",
+  "version": "1.0.18",
  "description": "TTS-related speech utilities for jambonz",
  "main": "index.js",
  "author": "Dave Horton",
@@ -13,7 +13,6 @@
    "coverage": "nyc --reporter html --report-dir ./coverage npm run test",
    "jslint": "eslint index.js lib",
    "jslint:fix": "npm run jslint --fix",
-    "build": "./build_stubs.sh",
    "prepare": "husky"
  },
  "repository": {
@@ -26,7 +25,6 @@
  },
  "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": "^2.2.7",
@@ -35,11 +33,9 @@
    "@jambonz/realtimedb-helpers": "^0.8.7",
    "bent": "^7.3.12",
    "debug": "^4.3.4",
-    "form-urlencoded": "^6.1.4",
    "google-protobuf": "^3.21.2",
-    "microsoft-cognitiveservices-speech-sdk": "1.38.0",
-    "openai": "^4.98.0",
-    "undici": "^7.5.0"
+    "microsoft-cognitiveservices-speech-sdk": "^1.51.0",
+    "openai": "^4.98.0"
  },
  "devDependencies": {
    "config": "^4.2.0",
@@ -1,332 +0,0 @@
-syntax = "proto3";
-
-package nuance.tts.v1;
-
-/*
-The Synthesizer service offers these functionalities:
-* - GetVoices: Queries the list of available voices, with filters to reduce the search space.  
-* - Synthesize: Synthesizes audio from input text and parameters, and returns an audio stream. 
-* - UnarySynthesize: Synthesizes audio from input text and parameters, and returns a single audio response. 
-*/
-service Synthesizer {
-  rpc GetVoices(GetVoicesRequest) returns (GetVoicesResponse) {}         
-  rpc Synthesize(SynthesisRequest) returns (stream SynthesisResponse) {} 
-  rpc UnarySynthesize(SynthesisRequest) returns (UnarySynthesisResponse) {}
-}
-
-/*
-Input message for [Synthesizer](#synthesizer) - GetVoices, to query voices available to the client.
-*/
-message GetVoicesRequest { 
-  Voice voice = 1; // Optionally filter the voices to retrieve, e.g. set language to en-US to return only American English voices.
-}
-
-/*
-Output message for [Synthesizer](#synthesizer) - GetVoices. Includes a list of voices that matched the input criteria, if any.
-*/
-message GetVoicesResponse { 
-  repeated Voice voices = 1;           // Repeated. Voices and characteristics returned.
-}
-
-/*
-Input message for [Synthesizer](#synthesizer) - Synthesize. Specifies input text, audio parameters, and events to subscribe to, in exchange for synthesized audio.
-*/
-message SynthesisRequest {
-  Voice voice = 1;                     // The voice to use for audio synthesis. Mandatory.
-  AudioParameters audio_params = 2;    // Output audio parameters, such as encoding and volume.
-  Input input = 3;                     // Input text to synthesize, tuning data, etc. Mandatory.
-  EventParameters event_params = 4;    // Markers and other info to include in server events returned during synthesis.
-  map<string, string> client_data = 5; // Repeated. Client-supplied key-value pairs to inject into the call log.
-  string user_id = 6;                  // Identifies a particular user within an application.
-}
-
-/*
-Input or output message for voices. When sent as input: 
-* - In [GetVoicesRequest](#getvoicesrequest), it filters the list of available voices.
-* - In [SynthesisRequest](#synthesisrequest), it specifies the voice to use for synthesis. 
-
-When received as output in [GetVoicesResponse](#getvoicesresponse), it returns the list of available voices.
-*/
-message Voice {
-  string name = 1;            // The voice's name, e.g. 'Evan'. Mandatory for SynthesisRequest. 
-  string model = 2;           // The voice's quality model, e.g. 'enhanced' or 'standard'. Mandatory for SynthesisRequest. 
-  string language = 3;        // IETF language code, e.g. 'en-US'. Used only for GetVoicesRequest and GetVoicesResponse, to search for voices with a certain mother tongue. Ignored otherwise. 
-  EnumAgeGroup age_group = 4; // Used only for GetVoicesRequest and GetVoicesResponse, to search for adult or child voices. Ignored otherwise.
-  EnumGender gender = 5;      // Used only for GetVoicesRequest and GetVoicesResponse, to search for voices with a certain gender. Ignored otherwise.
-  uint32 sample_rate_hz = 6;  // Used only for GetVoicesRequest and GetVoicesResponse, to search for a certain native sample rate. Ignored otherwise.
-  string language_tlw = 7;    // Used only for GetVoicesRequest and GetVoicesResponse. Three-letter language code (e.g. 'enu' for American English), for configuring language identification in [Input](#input).
-  bool restricted = 8;        // Used only in GetVoicesResponse, to identify restricted voices. These are custom voices available only to specific customers. Ignored otherwise.
-  string version = 9;         // Used only in GetVoicesResponse, to return the voice's version. Ignored otherwise.
-}
-
-/*
-Input or output field specifying whether the voice uses its adult or child version, if available. Included in [Voice](#voice). 
-*/
-enum EnumAgeGroup { 
-  ADULT = 0;      // Adult voice. Default for GetVoicesRequest.
-  CHILD = 1;      // Child voice. 
-}
-
-/*
-Input or output field, specifying gender for voices that support multiple genders. Included in [Voice](#voice). 
-*/
-enum EnumGender { 
-  ANY = 0;        // Any gender voice. Default for GetVoicesRequest. 
-  MALE = 1;       // Male voice.
-  FEMALE = 2;     // Female voice.
-  NEUTRAL = 3;    // Neutral gender voice. 
-}
-
-/*
-Input message for audio-related parameters during synthesis, including encoding, volume, and audio length. Included in [SynthesisRequest](#synthesisrequest). 
-*/
-message AudioParameters {
-  AudioFormat audio_format = 1;        // Audio encoding. Default PCM 22.5kHz. 
-  uint32 volume_percentage = 2;        // Volume amplitude, from 0 to 100. Default 80.
-  float speaking_rate_factor = 3;      // Speaking rate, from 0 to 2.0. Default 1.0.
-  uint32 audio_chunk_duration_ms = 4;  // Maximum duration, in ms, of an audio chunk delivered to the client, from 1 to 60000. Default is 20000 (20 seconds). When this parameter is large enough (for example, 20 or 30 seconds), each audio chunk contains an audible segment surrounded by silence.
-  uint32 target_audio_length_ms = 5;   // Maximum duration, in ms, of synthesized audio. When greater than 0, the server stops ongoing synthesis at the first sentence end, or silence, closest to the value.
-  bool disable_early_emission = 6;     // By default, audio segments are emitted as soon as possible, even if they are not audible. This behavior may be disabled.
-}
-
-/*
-Input message for audio encoding of synthesized text. Included in [AudioParameters](#audioparameters). 
-*/
-message AudioFormat {
-  oneof audio_format {
-    PCM pcm = 1;      // Signed 16-bit little endian PCM. 
-    ALaw alaw = 2;    // G.711 A-law, 8kHz.
-    ULaw ulaw = 3;    // G.711 Mu-law, 8kHz.
-    OggOpus ogg_opus = 4; // Ogg Opus, 8kHz, 16kHz or 24kHz.
-    Opus opus = 5;  // Opus, 8kHz, 16kHZ or 24 kHz. The audio will be sent one Opus packet at a time.
-  }
-}
-
-/* 
-Input message defining PCM sample rate. Included in [Audioformat](#audioformat).
-*/
-message PCM { 
-  uint32 sample_rate_hz = 1; // Output sample rate: 8000, 16000, 22050 (default), 24000. 
-}
-
-/*
-Input message defining A-law audio format. G.711 audio formats are set to 8kHz. Included in [Audioformat](#audioformat).
-*/
-message ALaw {}
-
-/*
-Input message defining Mu-law audio format. G.711 audio formats are set to 8kHz. Included in [Audioformat](#audioformat).
-*/
-message ULaw {}
-
-/*
-Input message defining Opus output rate. Included in [Audioformat](#audioformat).
-*/
-message Opus {
-  uint32 sample_rate_hz = 1;       // Output sample rate. Supported values: 8000, 16000, 24000 Hz. 
-  uint32 bit_rate_bps = 2;         // Valid range is 500 to 256000 bps. Default 28000 bps.
-  float max_frame_duration_ms = 3; // Opus frame size, in ms: 2.5, 5, 10, 20, 40, 60. Default 20.
-  uint32 complexity = 4;           // Computational complexity. A complexity of 0 means the codec default.
-  EnumVariableBitrate vbr = 5;     // Variable bitrate. On by default. 
-}
-
-/*
-Input message defining Ogg Opus output rate. Included in [Audioformat](#audioformat).
-*/
-message OggOpus {
-  uint32 sample_rate_hz = 1;       // Output sample rate. Supported values: 8000, 16000, 24000 Hz. 
-  uint32 bit_rate_bps = 2;         // Valid range is 500 to 256000 bps. Default 28000 bps.
-  float max_frame_duration_ms = 3; // Opus frame size, in ms: 2.5, 5, 10, 20, 40, 60. Default 20.
-  uint32 complexity = 4;           // Computational complexity. A complexity of 0 means the codec default.
-  EnumVariableBitrate vbr = 5;     // Variable bitrate. On by default. 
-}
-
-/*
-Settings for variable bitrate. Included in [OggOpus](#oggopus). Turned on by default.
-*/
-enum EnumVariableBitrate { 
-  VARIABLE_BITRATE_ON = 0;          // Use variable bitrate. Default. 
-  VARIABLE_BITRATE_OFF = 1;         // Do not use variable bitrate. 
-  VARIABLE_BITRATE_CONSTRAINED = 2; // Use constrained variable bitrate. 
-}
-
-/*
-Input message containing text to synthesize and synthesis parameters, including tuning data, etc. Included in [SynthesisRequest](#synthesisrequest). The type of input may be:
-* - Plain text.
-* - An SSML document.
-* - An alternating sequence of plain text and Nuance control codes.
-*/
-message Input {
-  oneof input_data {
-    Text text = 1;                            // Text input.  
-    SSML ssml = 2;                            // SSML input.
-    TokenizedSequence tokenized_sequence = 3; // Sequence of text and Nuance control codes.
-  }
-  repeated SynthesisResource resources = 4;        // Repeated. Synthesis resources (user dictionaries, rulesets, etc.) to tune synthesized audio. 
-  LanguageIdentificationParameters lid_params = 5; // LID parameters.  
-  DownloadParameters download_params = 6;          // Remote file download parameters.  
-}
-
-/*
-Input message for synthesizing plain text. The encoding must be in UTF-8.
-*/
-message Text {
-  oneof text_data {
-    string text = 1; // Plain input text in UTF-8 encoding.
-    string uri = 2;  // Remote URI to the plain input text. Disabled for Nuance-hosted TTS.
-  }
-}
-
-/*
-Input message for synthesizing an SSML document.
-*/
-message SSML {
-  oneof ssml_data {
-    string text = 1; // SSML input text.
-    string uri = 2;  // Remote URI to the SSML input text. Disabled for Nuance-hosted TTS.
-  }
-  EnumSSMLValidationMode ssml_validation_mode = 3;  // SSML validation mode. Default STRICT. 
-}
-
-/*
-Input message for synthesizing a sequence of plain text and Nuance control codes.
-*/
-message TokenizedSequence {
-  repeated Token tokens = 1;
-}
-
-/*
-The unit when using TokenizedSequence for input. Each token can either be plain text or a Nuance control code.
-*/
-message Token {
-  oneof token_data {
-    string text = 1; // Plain input text.
-    ControlCode control_code = 2; // Nuance control code.
-  }
-}
-
-/*
-A Nuance control code allows the user to control how text is spoken, similarly to SSML.
-*/
-message ControlCode {
-  string key = 1; // Name of the control code, e.g. "pause".
-  string value = 2; // Value of the control code.
-}
-
-/*
-Input message specifying the type of file to tune the synthesized output and its location or contents. Included in [Input](#input). 
-*/
-message SynthesisResource {
-  EnumResourceType type = 1;   // Resource type, e.g. user dictionary, etc. Default USER_DICTIONARY. 
-  oneof resource_data {
-    string uri = 2;            // URI to the remote resource, or
-    bytes body = 3;            // For EnumResourceType USER_DICTIONARY, the contents of the file.
-  }
-}
-
-/*
-The type of synthesis resource to tune the output. Included in [SynthesisResource](#synthesisresource). User dictionaries provide custom pronunciations, rulesets apply search-and-replace rules to input text and ActivePrompt databases help tune synthesized audio under certain conditions, using Nuance Vocalizer Studio. 
-*/
-enum EnumResourceType {
-  USER_DICTIONARY = 0;       // User dictionary (application/edct-bin-dictionary). Default. 
-  TEXT_USER_RULESET = 1;     // Text user ruleset (application/x-vocalizer-rettt+text).
-  BINARY_USER_RULESET = 2;   // Binary user ruleset (application/x-vocalizer-rettt+bin).
-  ACTIVEPROMPT_DB = 3;       // ActivePrompt database (application/x-vocalizer/activeprompt-db).
-  ACTIVEPROMPT_DB_AUTO = 4;  // ActivePrompt database with automatic insertion (application/x-vocalizer/activeprompt-db;mode=automatic).
-  SYSTEM_DICTIONARY = 5;     // Nuance system dictionary (application/sdct-bin-dictionary).
-}
-
-/*
-SSML validation mode when using SSML input. Included in [Input](#input). Strict by default but can be relaxed.
-*/
-enum EnumSSMLValidationMode { 
-  STRICT = 0;   // Strict SSL validation. Default. 
-  WARN = 1;     // Give warning only. 
-  NONE = 2;     // Do not validate. 
-}
-
-/*
-Input message controlling the language identifier. Included in [Input](#input). The language identifier runs on input blocks labeled with the <ESC>\lang=unknown\ control sequence or SSML xml:lang="unknown". The language identifier automatically restricts the matched languages to the installed voices. This limits the permissible languages, and also sets the order of precedence (first to last) when they have equal confidence scores.
-*/
-message LanguageIdentificationParameters {
-  bool disable = 1;                       // Whether to disable language identification. Turned on by default. 
-  repeated string languages = 2;          // Repeated. List of three-letter language codes (e.g. enu, frc, spm) to restrict language identification results, in order of precedence. Use GetVoicesRequest - Voice - language_tlw to obtain the three-letter codes. Default blank. 
-  bool always_use_highest_confidence = 3; // If enabled, language identification always chooses the language with the highest confidence score, even if the score is low. Default false, meaning use language with any confidence.
-}
-
-/*
-Input message containing parameters for remote file download, whether for input text (Input.uri) or a SynthesisResource (SynthesisResource.uri). Included in [Input](#input). 
-*/
-message DownloadParameters {
-  map <string,string> headers = 1; // HTTP headers to include in outgoing requests. Only whitelisted headers will actually be sent.
-  bool refuse_cookies = 2;         // Whether to disable cookies. By default, HTTP requests accept cookies.
-  oneof optional_download_parameter_request_timeout_ms {
-    uint32 request_timeout_ms = 3;       // Request timeout in ms. Default (0) means server default, usually 30000 (30 seconds).
-  }
-}
-
-/*
-Input message that defines event subscription parameters. Included in [SynthesisRequest](#synthesisrequest). Events that are requested are sent throughout the SynthesisResponse stream, when generated.  Marker events can send events as certain parts of the synthesized audio are reached, for example, at the end of a word, sentence, or user-defined bookmark.
-
-* Log events are produced throughout a synthesis request for events such as a voice loaded by the server or an audio chunk being ready to send.
-*/
-message EventParameters {
-  bool send_sentence_marker_events = 1;   // Sentence marker. Default: do not send. 
-  bool send_word_marker_events = 2;       // Word marker. Default: do not send. 
-  bool send_phoneme_marker_events = 3;    // Phoneme marker. Default: do not send. 
-  bool send_bookmark_marker_events = 4;   // Bookmark marker. Default: do not send. 
-  bool send_paragraph_marker_events = 5;  // Paragraph marker. Default: do not send. 
-  bool send_visemes = 6;                  // Lipsync information. Default: do not send. 
-  bool send_log_events = 7;               // Whether to log events during synthesis. By default, logging is turned off.
-  bool suppress_input = 8;                // Whether to omit input text and URIs from log events. By default, these items are included.
-}
-
-/*
-The [Synthesizer](#synthesizer) - Synthesize RPC call returns a stream of SynthesisResponse messages. The response contains one of: 
-* - A status response, indicating completion or failure of the request. This is received only once and signifies the end of a Synthesize call.
-* - A list of events the client has requested. This can be received many times. See EventParameters for details.
-* - An audio buffer. This may be received many times.
-*/
-message SynthesisResponse {
-  oneof response {
-    Status status = 1;   // A status response, indicating completion or failure of the request.
-    Events events = 2;   // A list of events. See EventParameters for details.
-    bytes audio = 3;     // The latest audio buffer.
-  }
-}
-
-/*
-The [Synthesizer](#synthesizer) - UnarySynthesize RPC call returns a single UnarySynthesisResponse message. It is similar to a SynthesisResponse message but includes all the information instead of a single type of response. The response contains: 
-* - A status response, indicating completion or failure of the request. 
-* - A list of events the client has requested. See EventParameters for details.
-* - The complete audio buffer of the synthesized text.
-*/
-message UnarySynthesisResponse {
-  Status status = 1;   // A status response, indicating completion or failure of the request.
-  Events events = 2;   // A list of events. See EventParameters for details.
-  bytes audio = 3;     // Audio buffer of the synthesized text. 
-}
-
-/*
-Output message containing a status response, indicating completion or failure of a SynthesisRequest. Included in [SynthesisResponse](#synthesisresponse) and [UnarySynthesisResponse](#unarysynthesisresponse).
-*/
-message Status {
-  uint32 code = 1;    // HTTP-style return code: 200, 4xx, or 5xx as appropriate.
-  string message = 2; // Brief description of the status.
-  string details = 3; // Longer description if available.
-}
-
-/*
-Output message defining a container for a list of events. This container is needed because oneof does not allow repeated parameters in Protobuf.  Included in [SynthesisResponse](#synthesisresponse) and [UnarySynthesisResponse](#unarysynthesisresponse).
-*/
-message Events {
-  repeated Event events = 1;      // Repeated. One or more events. 
-}
-
-/*
-Output message defining an event message.  Included in [Events](#events). See EventParameters for details.
-*/
-message Event {
-  string name = 1;                // Either "Markers" or the name of the event in the case of a Log Event.
-  map<string, string> values = 2; // Repeated. Key-value data relevant to the current event.
-}
@@ -1,104 +0,0 @@
-// GENERATED CODE -- DO NOT EDIT!
-
-'use strict';
-var grpc = require('@grpc/grpc-js');
-var synthesizer_pb = require('./synthesizer_pb.js');
-
-function serialize_nuance_tts_v1_GetVoicesRequest(arg) {
-  if (!(arg instanceof synthesizer_pb.GetVoicesRequest)) {
-    throw new Error('Expected argument of type nuance.tts.v1.GetVoicesRequest');
-  }
-  return Buffer.from(arg.serializeBinary());
-}
-
-function deserialize_nuance_tts_v1_GetVoicesRequest(buffer_arg) {
-  return synthesizer_pb.GetVoicesRequest.deserializeBinary(new Uint8Array(buffer_arg));
-}
-
-function serialize_nuance_tts_v1_GetVoicesResponse(arg) {
-  if (!(arg instanceof synthesizer_pb.GetVoicesResponse)) {
-    throw new Error('Expected argument of type nuance.tts.v1.GetVoicesResponse');
-  }
-  return Buffer.from(arg.serializeBinary());
-}
-
-function deserialize_nuance_tts_v1_GetVoicesResponse(buffer_arg) {
-  return synthesizer_pb.GetVoicesResponse.deserializeBinary(new Uint8Array(buffer_arg));
-}
-
-function serialize_nuance_tts_v1_SynthesisRequest(arg) {
-  if (!(arg instanceof synthesizer_pb.SynthesisRequest)) {
-    throw new Error('Expected argument of type nuance.tts.v1.SynthesisRequest');
-  }
-  return Buffer.from(arg.serializeBinary());
-}
-
-function deserialize_nuance_tts_v1_SynthesisRequest(buffer_arg) {
-  return synthesizer_pb.SynthesisRequest.deserializeBinary(new Uint8Array(buffer_arg));
-}
-
-function serialize_nuance_tts_v1_SynthesisResponse(arg) {
-  if (!(arg instanceof synthesizer_pb.SynthesisResponse)) {
-    throw new Error('Expected argument of type nuance.tts.v1.SynthesisResponse');
-  }
-  return Buffer.from(arg.serializeBinary());
-}
-
-function deserialize_nuance_tts_v1_SynthesisResponse(buffer_arg) {
-  return synthesizer_pb.SynthesisResponse.deserializeBinary(new Uint8Array(buffer_arg));
-}
-
-function serialize_nuance_tts_v1_UnarySynthesisResponse(arg) {
-  if (!(arg instanceof synthesizer_pb.UnarySynthesisResponse)) {
-    throw new Error('Expected argument of type nuance.tts.v1.UnarySynthesisResponse');
-  }
-  return Buffer.from(arg.serializeBinary());
-}
-
-function deserialize_nuance_tts_v1_UnarySynthesisResponse(buffer_arg) {
-  return synthesizer_pb.UnarySynthesisResponse.deserializeBinary(new Uint8Array(buffer_arg));
-}
-
-
-//
-// The Synthesizer service offers these functionalities:
-// - GetVoices: Queries the list of available voices, with filters to reduce the search space.
-// - Synthesize: Synthesizes audio from input text and parameters, and returns an audio stream.
-// - UnarySynthesize: Synthesizes audio from input text and parameters, and returns a single audio response.
-var SynthesizerService = exports.SynthesizerService = {
-  getVoices: {
-    path: '/nuance.tts.v1.Synthesizer/GetVoices',
-    requestStream: false,
-    responseStream: false,
-    requestType: synthesizer_pb.GetVoicesRequest,
-    responseType: synthesizer_pb.GetVoicesResponse,
-    requestSerialize: serialize_nuance_tts_v1_GetVoicesRequest,
-    requestDeserialize: deserialize_nuance_tts_v1_GetVoicesRequest,
-    responseSerialize: serialize_nuance_tts_v1_GetVoicesResponse,
-    responseDeserialize: deserialize_nuance_tts_v1_GetVoicesResponse,
-  },
-  synthesize: {
-    path: '/nuance.tts.v1.Synthesizer/Synthesize',
-    requestStream: false,
-    responseStream: true,
-    requestType: synthesizer_pb.SynthesisRequest,
-    responseType: synthesizer_pb.SynthesisResponse,
-    requestSerialize: serialize_nuance_tts_v1_SynthesisRequest,
-    requestDeserialize: deserialize_nuance_tts_v1_SynthesisRequest,
-    responseSerialize: serialize_nuance_tts_v1_SynthesisResponse,
-    responseDeserialize: deserialize_nuance_tts_v1_SynthesisResponse,
-  },
-  unarySynthesize: {
-    path: '/nuance.tts.v1.Synthesizer/UnarySynthesize',
-    requestStream: false,
-    responseStream: false,
-    requestType: synthesizer_pb.SynthesisRequest,
-    responseType: synthesizer_pb.UnarySynthesisResponse,
-    requestSerialize: serialize_nuance_tts_v1_SynthesisRequest,
-    requestDeserialize: deserialize_nuance_tts_v1_SynthesisRequest,
-    responseSerialize: serialize_nuance_tts_v1_UnarySynthesisResponse,
-    responseDeserialize: deserialize_nuance_tts_v1_UnarySynthesisResponse,
-  },
-};
-
-exports.SynthesizerClient = grpc.makeGenericClientConstructor(SynthesizerService);
@@ -3,5 +3,4 @@ require('./synth');
 require('./list-voices');
 require('./aws');

-require('./nuance');
 require('./docker_stop');
@@ -12,99 +12,6 @@ const stats = {
  histogram: () => {}
 };

-test('Verbio - get Access key and voices', async(t) => {
-  const fn = require('..');
-  const {client, getTtsVoices, getVerbioAccessToken} = fn(opts, logger);
-  if (!process.env.VERBIO_CLIENT_ID || !process.env.VERBIO_CLIENT_SECRET) {
-    t.pass('skipping Verbio test since no Verbio Keys provided');
-    t.end();
-    client.quit();
-    return;
-  }
-
-  try {
-    const credentials = {
-      client_id: process.env.VERBIO_CLIENT_ID,
-      client_secret: process.env.VERBIO_CLIENT_SECRET
-    };
-    let obj = await getVerbioAccessToken(credentials);
-    t.ok(obj.access_token , 'successfully received access token not from cache');
-    const voices = await getTtsVoices({vendor: 'verbio', credentials});
-    t.ok(voices && voices.length != 0, 'successfully received verbio voices');
-  } catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
-test('Nuance hosted tests', async(t) => {
-  const fn = require('..');
-  const {client, getTtsVoices} = fn(opts, logger);
-
-  if (!process.env.NUANCE_CLIENT_ID || !process.env.NUANCE_SECRET ) {
-      t.pass('skipping Nuance hosted test since no Nuance client_id and secret provided');
-      t.end();
-      client.quit();
-      return;
-  }
-  try {
-    const opts = {
-      vendor: 'nuance',
-      credentials: {
-        client_id: process.env.NUANCE_CLIENT_ID,
-        secret: process.env.NUANCE_SECRET
-      }
-    };
-    let voices = await getTtsVoices(opts);
-    t.ok(voices.length > 0 && voices[0].language, 
-      `GetVoices: successfully retrieved ${voices.length} voices from Nuance`);
-
-    await client.flushall();
-
-    t.end();
-
-  }
-  catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
-test('Nuance on-prem tests', async(t) => {
-  const fn = require('..');
-  const {client, getTtsVoices} = fn(opts, logger);
-
-  if (!process.env.NUANCE_TTS_URI ) {
-      t.pass('skipping Nuance on-prem test since no Nuance uri provided');
-      t.end();
-      client.quit();
-      return;
-  }
-  try {
-    const opts = {
-      vendor: 'nuance',
-      credentials: {
-        nuance_tts_uri: process.env.NUANCE_TTS_URI
-      }
-    };
-    let voices = await getTtsVoices(opts);
-    t.ok(voices.length > 0 && voices[0].language, 
-      `GetVoices: successfully retrieved ${voices.length} voices from Nuance`);
-
-    await client.flushall();
-
-    t.end();
-
-  }
-  catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
 test('Google tests', async(t) => {
  const fn = require('..');
  const {client, getTtsVoices} = fn(opts, logger);
@@ -1,81 +0,0 @@
-const test = require('tape').test ;
-const config = require('config');
-const opts = config.get('redis');
-const fs = require('fs');
-const logger = require('pino')({level: 'error'});
-process.on('unhandledRejection', (reason, p) => {
-  console.log('Unhandled Rejection at: Promise', p, 'reason:', reason);
-});
-
-const stats = {
-  increment: () => {},
-  histogram: () => {}
-};
-
-test('Nuance hosted tests', async(t) => {
-  const fn = require('..');
-  const {client, getTtsVoices} = fn(opts, logger);
-
-  if (!process.env.NUANCE_CLIENT_ID || !process.env.NUANCE_SECRET ) {
-      t.pass('skipping Nuance hosted test since no Nuance client_id and secret provided');
-      t.end();
-      client.quit();
-      return;
-  }
-  try {
-    const opts = {
-      vendor: 'nuance',
-      credentials: {
-        client_id: process.env.NUANCE_CLIENT_ID,
-        secret: process.env.NUANCE_SECRET
-      }
-    };
-    let voices = await getTtsVoices(opts);
-    t.ok(voices.length > 0 && voices[0].language, 
-      `GetVoices: successfully retrieved ${voices.length} voices from Nuance`);
-
-    await client.flushall();
-
-    t.end();
-
-  }
-  catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
-test('Nuance on-prem tests', async(t) => {
-  const fn = require('..');
-  const {client, getTtsVoices} = fn(opts, logger);
-
-  if (!process.env.NUANCE_TTS_URI ) {
-      t.pass('skipping Nuance on-prem test since no Nuance uri provided');
-      t.end();
-      client.quit();
-      return;
-  }
-  try {
-    const opts = {
-      vendor: 'nuance',
-      credentials: {
-        nuance_tts_uri: process.env.NUANCE_TTS_URI
-      }
-    };
-    let voices = await getTtsVoices(opts);
-    t.ok(voices.length > 0 && voices[0].language, 
-      `GetVoices: successfully retrieved ${voices.length} voices from Nuance`);
-
-    await client.flushall();
-
-    t.end();
-
-  }
-  catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
@@ -442,6 +442,95 @@ test('Google TTS streaming tests (!JAMBONES_DISABLE_TTS_STREAMING)', async(t) =>
    });
    t.ok(result.filePath.includes('api_mode=tts'), 'options.apiMode=tts overrides HD voice default');

+    /* AudioConfig settings (speakingRate, pitch, volumeGainDb) */
+    const googleCreds = {
+      credentials: {
+        client_email: creds.client_email,
+        private_key: creds.private_key,
+      },
+    };
+
+    // Test 9: AudioConfig nested under options.audioConfig, as in google's API docs
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Wavenet-D',
+      text: 'Testing nested audioConfig settings.',
+      options: { audioConfig: { speakingRate: 1.4, pitch: -2.5, volumeGainDb: 6 } },
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.includes(',speaking_rate=1.4'), 'nested audioConfig sets speaking_rate');
+    t.ok(result.filePath.includes(',pitch=-2.5'), 'nested audioConfig sets pitch');
+    t.ok(result.filePath.includes(',volume_gain_db=6'), 'nested audioConfig sets volume_gain_db');
+
+    // Test 10: AudioConfig flat at the top level of options, in camelCase or snake_case
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Wavenet-D',
+      text: 'Testing flat audioConfig settings.',
+      options: { speaking_rate: '0.8', volumeGainDb: 3 },
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.includes(',speaking_rate=0.8'), 'flat snake_case speaking_rate is honored');
+    t.ok(result.filePath.includes(',volume_gain_db=3'), 'flat camelCase volumeGainDb is honored');
+    t.ok(!result.filePath.includes(',pitch='), 'unspecified audioConfig setting is omitted');
+
+    // Test 11: zero is a meaningful value for pitch and volumeGainDb, not an absent one
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Wavenet-D',
+      text: 'Testing zero audioConfig settings.',
+      options: { audioConfig: { pitch: 0, volumeGainDb: 0 } },
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.includes(',pitch=0'), 'pitch=0 is passed through rather than dropped');
+    t.ok(result.filePath.includes(',volume_gain_db=0'), 'volume_gain_db=0 is passed through rather than dropped');
+
+    // Test 12: non-numeric values are ignored, so they cannot corrupt the freeswitch param string
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Wavenet-D',
+      text: 'Testing invalid audioConfig settings.',
+      options: { audioConfig: { speakingRate: 'fast,evil=1', pitch: null, volumeGainDb: '' } },
+      disableTtsCache: true
+    });
+    t.ok(!result.filePath.includes('speaking_rate'), 'non-numeric speakingRate is ignored');
+    t.ok(!result.filePath.includes('evil=1'), 'non-numeric value cannot inject extra params');
+    t.ok(!result.filePath.includes('pitch='), 'null pitch is ignored');
+    t.ok(!result.filePath.includes('volume_gain_db'), 'empty volumeGainDb is ignored');
+
+    // Test 13: no audioConfig supplied leaves the param string untouched
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Wavenet-D',
+      text: 'Testing absent audioConfig settings.',
+      disableTtsCache: true
+    });
+    t.ok(!/speaking_rate|pitch=|volume_gain_db/.test(result.filePath),
+      'no audioConfig params are added when none are supplied');
+
+    // Test 14: HD voice (api_mode=live) also carries speaking_rate, the one setting google streams support
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Chirp3-HD-Charon',
+      text: 'Testing audioConfig on an HD voice.',
+      options: { audioConfig: { speakingRate: 1.25 } },
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.includes('api_mode=live'), 'HD voice with audioConfig still uses api_mode=live');
+    t.ok(result.filePath.includes(',speaking_rate=1.25'), 'HD voice streaming path carries speaking_rate');
+
  } catch (err) {
    console.error(err);
    t.end(err);
@@ -526,6 +615,42 @@ test('Google TTS non-streaming tests (JAMBONES_DISABLE_TTS_STREAMING=true)', asy
    t.ok(!result.filePath.startsWith('say:'), 'Gemini TTS does NOT return streaming say: path when disabled');
    t.ok(result.filePath.endsWith('.mp3'), 'Gemini TTS returns mp3 file path');

+    const googleCreds = {
+      credentials: {
+        client_email: creds.client_email,
+        private_key: creds.private_key,
+      },
+    };
+
+    /**
+     * Test 4: AudioConfig settings are accepted by the synthesize API.
+     * Google rejects out-of-range values with a 400, so a successful render also confirms
+     * the settings reached the request rather than being silently dropped.
+     */
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'en-US-Wavenet-D',
+      text: 'This is a test of audioConfig with streaming disabled.',
+      options: { audioConfig: { speakingRate: 1.4, pitch: -2.5, volumeGainDb: 6 } },
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.endsWith('.mp3'), 'standard voice renders mp3 with audioConfig settings applied');
+
+    /* Test 5: gemini voices ignore the AudioConfig settings rather than failing on them */
+    result = await synthAudio(stats, {
+      vendor: 'google',
+      credentials: googleCreds,
+      language: 'en-US',
+      voice: 'Kore',
+      model: geminiModel,
+      text: 'This is a test of audioConfig on Gemini TTS.',
+      options: { audioConfig: { speakingRate: 1.4, pitch: -2.5, volumeGainDb: 6 } },
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.endsWith('.mp3'), 'gemini voice renders mp3 with audioConfig settings skipped');
+
  } catch (err) {
    console.error(err);
    t.end(err);
@@ -759,82 +884,6 @@ test('Azure custom voice speech synth tests', async(t) => {
  client.quit();
 });

-test('Nuance hosted speech synth tests', async(t) => {
-  const fn = require('..');
-  const {synthAudio, client} = fn(opts, logger);
-
-  if (!process.env.NUANCE_CLIENT_ID || !process.env.NUANCE_SECRET) {
-    t.pass('skipping Nuance speech synth tests since NUANCE_CLIENT_ID or NUANCE_SECRET not provided');
-    return t.end();
-  }
-  try {
-    let opts = await synthAudio(stats, {
-      vendor: 'nuance',
-      credentials: {
-        client_id: process.env.NUANCE_CLIENT_ID,
-        secret: process.env.NUANCE_SECRET,
-      },
-      language: 'en-US',
-      voice: 'Evan',
-      text: 'This is a test.  This is only a test',
-    });
-    t.ok(!opts.servedFromCache, `successfully synthesized nuance audio to ${opts.filePath}`);
-
-    opts = await synthAudio(stats, {
-      vendor: 'nuance',
-      credentials: {
-        client_id: process.env.NUANCE_CLIENT_ID,
-        secret: process.env.NUANCE_SECRET,
-      },
-      language: 'en-US',
-      voice: 'Evan',
-      text: 'This is a test.  This is only a test',
-    });
-    t.ok(opts.servedFromCache, `successfully retrieved nuance audio from cache ${opts.filePath}`);
-  } catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
-test('Nuance on-prem speech synth tests', async(t) => {
-  const fn = require('..');
-  const {synthAudio, client} = fn(opts, logger);
-
-  if (!process.env.NUANCE_TTS_URI) {
-    t.pass('skipping Nuance on prem speech synth tests since NUANCE_TTS_URI not provided');
-    return t.end();
-  }
-  try {
-    let opts = await synthAudio(stats, {
-      vendor: 'nuance',
-      credentials: {
-        nuance_tts_uri: process.env.NUANCE_TTS_URI
-      },
-      language: 'en-US',
-      voice: 'Evan',
-      text: 'This is a test of on-prem.  This is only a test',
-    });
-    t.ok(!opts.servedFromCache, `successfully synthesized nuance audio to ${opts.filePath}`);
-
-    opts = await synthAudio(stats, {
-      vendor: 'nuance',
-      credentials: {
-        nuance_tts_uri: process.env.NUANCE_TTS_URI
-      },
-      language: 'en-US',
-      voice: 'Evan',
-      text: 'This is a test of on-prem.  This is only a test',
-    });
-    t.ok(opts.servedFromCache, `successfully retrieved nuance audio from cache ${opts.filePath}`);
-  } catch (err) {
-    console.error(err);
-    t.end(err);
-  }
-  client.quit();
-});
-
 test('Nvidia speech synth tests', async(t) => {
  const fn = require('..');
  const {synthAudio, client} = fn(opts, logger);
@@ -975,55 +1024,6 @@ test('Elevenlabs speech synth tests', async(t) => {
  client.quit();
 });

-const testPlayHT = async(t, voice_engine) => {
-  const fn = require('..');
-  const {synthAudio, client} = fn(opts, logger);
-
-  if (!process.env.PLAYHT_API_KEY || !process.env.PLAYHT_USER_ID) {
-    t.pass('skipping PlayHT speech synth tests since PLAYHT_API_KEY or PLAYHT_USER_ID is/are not provided');
-    return t.end();
-  }
-  const text = 'Hi there and welcome to jambones! ' + Date.now();
-  try {
-    const opts = await synthAudio(stats, {
-      vendor: 'playht',
-      credentials: {
-        api_key: process.env.PLAYHT_API_KEY,
-        user_id: process.env.PLAYHT_USER_ID,
-        voice_engine,
-        options: JSON.stringify({
-          quality: 'medium',
-          speed: 1,
-          seed: 1,
-          temperature: 1,
-          emotion: 'female_happy',
-          voice_guidance: 3,
-          style_guidance: 20,
-          text_guidance: 1,
-        })
-      },
-      language: 'english',
-      voice: 's3://voice-cloning-zero-shot/d9ff78ba-d016-47f6-b0ef-dd630f59414e/female-cs/manifest.json',
-      text,
-      renderForCaching: true
-    });
-    t.ok(!opts.servedFromCache, `successfully playht eleven audio to ${opts.filePath}`);
-
-  } catch (err) {
-    console.error(JSON.stringify(err));
-    t.end(err);
-  }
-  client.quit();
-};
-
-test('PlayHT speech synth tests', async(t) => {
-  await testPlayHT(t, 'PlayHT2.0-turbo');
-});
-
-test('PlayHT3.0 speech synth tests', async(t) => {
-  await testPlayHT(t, 'Play3.0');
-});
-
 test('Cartesia speech synth tests', async(t) => {
  const fn = require('..');
  const {synthAudio, client} = fn(opts, logger);
@@ -1058,6 +1058,106 @@ test('Cartesia speech synth tests', async(t) => {
  client.quit();
 });

+test('gradium speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.GRADIUM_API_KEY) {
+    t.pass('skipping gradium speech synth tests since GRADIUM_API_KEY is not provided');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambones! ' + Date.now();
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'gradium',
+      credentials: {
+        api_key: process.env.GRADIUM_API_KEY,
+        model_id: 'default'
+      },
+      voice: 'YTpq7expH9539ERJ',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache, `successfully synthed gradium audio to ${opts.filePath}`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
+test('fishaudio speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.FISHAUDIO_API_KEY) {
+    t.pass('skipping fishaudio speech synth tests since FISHAUDIO_API_KEY is not provided');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambones! ' + Date.now();
+  try {
+    /* voice 'default' means "send no reference_id" — fish's own default voice */
+    const o = await synthAudio(stats, {
+      vendor: 'fishaudio',
+      credentials: {
+        api_key: process.env.FISHAUDIO_API_KEY,
+        model_id: 's2.1-pro'
+      },
+      voice: 'default',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!o.servedFromCache, `successfully synthed fishaudio audio to ${o.filePath}`);
+
+    /* the cache render must be raw 8k pcm (r8): fish's wav header carries a
+       placeholder RIFF size, so we never ask for wav */
+    const o2 = await synthAudio(stats, {
+      vendor: 'fishaudio',
+      credentials: {api_key: process.env.FISHAUDIO_API_KEY},
+      voice: 'default',
+      text: text + ' two',
+      renderForCaching: true,
+      disableTtsCache: true
+    });
+    t.ok(!o2.servedFromCache, 'fishaudio synthed a second uncached render');
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
+test('nineninesix speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.NINENINESIX_API_KEY) {
+    t.pass('skipping nineninesix speech synth tests since NINENINESIX_API_KEY is not provided');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambones! ' + Date.now();
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'nineninesix',
+      credentials: {
+        api_key: process.env.NINENINESIX_API_KEY,
+        model_id: 'gepard-1.0'
+      },
+      language: 'en',
+      voice: '775e4dfd-819c-4325-92ec-250f487ef7e3',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache, `successfully synthed nineninesix audio to ${opts.filePath}`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
 test('inworld speech synth', async(t) => {
  const fn = require('..');
  const {synthAudio, client} = fn(opts, logger);
@@ -1088,6 +1188,69 @@ test('inworld speech synth', async(t) => {
  client.quit();
 });

+test('inworld streaming say: params', async(t) => {
+  /* This test asserts the streaming say: path, so it must run with streaming
+     enabled. The Google non-streaming test above sets
+     JAMBONES_DISABLE_TTS_STREAMING and, on its no-credentials skip path,
+     deletes the env var WITHOUT clearing the require cache — so lib/config can
+     still be holding 'true' by the time we get here. Re-require to be
+     independent of what ran before us.
+   */
+  delete process.env.JAMBONES_DISABLE_TTS_STREAMING;
+  delete require.cache[require.resolve('../lib/config')];
+  delete require.cache[require.resolve('../lib/synth-audio')];
+  delete require.cache[require.resolve('..')];
+
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.INWORLD_API_KEY) {
+    t.pass('skipping inworld streaming say: param tests since INWORLD_API_KEY is not provided');
+    client.quit();
+    return t.end();
+  }
+
+  try {
+    let result = await synthAudio(stats, {
+      vendor: 'inworld',
+      credentials: {api_key: process.env.INWORLD_API_KEY, model_id: 'inworld-tts-1.5-mini'},
+      language: 'en',
+      voice: 'Ashley',
+      text: 'This is a test of inworld streaming.',
+      options: {temperature: 0.9, audioConfig: {pitch: 2.5, speakingRate: 1.2}},
+      disableTtsCache: true
+    });
+    t.ok(result.filePath.startsWith('say:'), 'inworld returns streaming say: path');
+    t.ok(result.filePath.includes('vendor=inworld'), 'streaming path contains vendor=inworld');
+    t.ok(result.filePath.includes('voice=Ashley'), 'streaming path contains voice');
+    t.ok(result.filePath.includes('model_id=inworld-tts-1.5-mini'), 'streaming path contains model_id');
+    t.ok(result.filePath.includes('temperature=0.9'), 'streaming path contains temperature');
+    /* pitch and speakingRate are nested under audioConfig; they used to be read
+       from the top level and emitted as "undefined"
+     */
+    t.ok(result.filePath.includes('speakingRate=1.2'), 'audioConfig.speakingRate reaches the say: params');
+    t.ok(result.filePath.includes('pitch=2.5'), 'audioConfig.pitch reaches the say: params');
+    t.ok(!result.filePath.includes('undefined'), 'no undefined values in the say: params');
+
+    /* options omitted entirely: no stray keys */
+    result = await synthAudio(stats, {
+      vendor: 'inworld',
+      credentials: {api_key: process.env.INWORLD_API_KEY, model_id: 'inworld-tts-1.5-mini'},
+      language: 'en',
+      voice: 'Ashley',
+      text: 'This is a test of inworld streaming.',
+      disableTtsCache: true
+    });
+    t.ok(!result.filePath.includes('speakingRate='), 'speakingRate omitted when unset');
+    t.ok(!result.filePath.includes('pitch='), 'pitch omitted when unset');
+    t.ok(!result.filePath.includes('undefined'), 'no undefined values when options are omitted');
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
 test('resemble speech synth', async(t) => {
  const fn = require('..');
  const {synthAudio, client} = fn(opts, logger);
@@ -1215,39 +1378,6 @@ test('whisper speech synth tests', async(t) => {
  client.quit();
 });

-test('Verbio speech synth tests', async(t) => {
-  const fn = require('..');
-  const {synthAudio, client} = fn(opts, logger);
-
-  if (!process.env.VERBIO_CLIENT_ID || !process.env.VERBIO_CLIENT_SECRET) {
-    t.pass('skipping Verbio Synthesize test since no Verbio Keys provided');
-    t.end();
-    client.quit();
-    return;
-  }
-
-  const text = 'Hi there and welcome to jambones!';
-  try {
-    let opts = await synthAudio(stats, {
-      vendor: 'verbio',
-      credentials: {
-        client_id: process.env.VERBIO_CLIENT_ID,
-        client_secret: process.env.VERBIO_CLIENT_SECRET
-      },
-      language: 'en-US',
-      voice: 'tommy_en-us',
-      text,
-      renderForCaching: true
-    });
-    t.ok(!opts.servedFromCache, `successfully synthesized whisper audio to ${opts.filePath}`);
-
-  } catch (err) {
-    console.error(JSON.stringify(err));
-    t.end(err);
-  }
-  client.quit();
-})
-
 test('Deepgram speech synth tests', async(t) => {
  const fn = require('..');
  const {synthAudio, client} = fn(opts, logger);
@@ -1276,6 +1406,69 @@ test('Deepgram speech synth tests', async(t) => {
  client.quit();
 })

+test('Deepgram Flux speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.DEEPGRAM_API_KEY) {
+    t.pass('skipping Deepgram Flux speech synth tests since DEEPGRAM_API_KEY');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambones!';
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'deepgramflux',
+      credentials: {
+        api_key: process.env.DEEPGRAM_API_KEY
+      },
+      model: process.env.DEEPGRAM_FLUX_MODEL || 'flux-alexis-en',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache, `successfully synthesized deepgramflux audio to ${opts.filePath}`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
+test('xai speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.XAI_API_KEY) {
+    t.pass('skipping xai speech synth tests - no XAI_API_KEY');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambones!';
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'xai',
+      credentials: {
+        api_key: process.env.XAI_API_KEY,
+        options: JSON.stringify({
+          voice: process.env.XAI_VOICE || 'eve',
+          speed: 1.0,
+          optimize_streaming_latency: 1,
+          text_normalization: true
+        })
+      },
+      language: 'en',
+      voice: process.env.XAI_VOICE || 'eve',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache && opts.filePath, `successfully synthesized xai audio to ${opts.filePath}`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
 test('TTS Cache tests', async(t) => {
  const fn = require('..');
  const {purgeTtsCache, getTtsSize, client} = fn(opts, logger);
Author	SHA1	Message	Date
Dave Horton	f4c3c7dc8b	1.0.18	2026-08-23 13:12:37 -04:00
Dave HortonandClaude Opus 5	a32f71ac5a	feat: add fishaudio (Fish Audio) TTS support (#159 ) Adds synthFishaudio with both arms: the say: streaming url consumed by the mediajam dialect, and a POST /v1/tts cache render. The render asks for raw pcm at 8k and returns extension r8 because fish's wav output carries a placeholder RIFF size, the same problem gradium has. Fish is a voice-cloning vendor, so the voice is a reference_id; the sentinel 'default' means send none and use fish's own default voice. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-23 13:11:59 -04:00
Dave HortonandClaude Opus 5	c8850998b5	fix(tts): use a current nineninesix voice id in the synth test (#158 ) The vendor replaced its voice catalog and now rejects unknown ids, so the env-gated test failed whenever NINENINESIX_API_KEY was set. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-16 12:19:59 -04:00
Dave Horton	e8b2009d29	1.0.17	2026-08-10 16:45:32 -04:00
Dave HortonandClaude Opus 5	27e07bb551	fix(tts): forward gradium json_config to the streaming path (#157 ) synthGradium destructured json_config from options but only forwarded pronunciation_id into the say: param block, so the setting could never reach mediajam's streaming dialect — only the cache-render POST. The say: parser is brace-aware, so a nested json object survives intact. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-10 16:45:07 -04:00
Dave Horton	33456b93d9	1.0.16	2026-08-08 23:07:43 -04:00
Dave HortonandClaude Opus 5	70e91b5057	test(inworld): gate the say: param test on INWORLD_API_KEY (#156 ) The test I added in #155 ran unconditionally, which broke `npm test` without credentials — the path husky's pre-commit hook takes, so `npm version patch` could not commit. Two causes, both addressed: - no credential gate, unlike every other vendor test in this file. Now skips without INWORLD_API_KEY, and closes its redis client on that path so the run can still exit. - it assumed streaming was enabled. The Google non-streaming test sets JAMBONES_DISABLE_TTS_STREAMING and, on its no-credentials skip path, deletes the env var WITHOUT clearing the require cache (unlike its finally block, which clears both) — so lib/config still held 'true' further down the file and synthInworld took the non-streaming branch, attempting a real vendor call. The test now re-requires with streaming enabled so it does not depend on what ran before it. Verified both ways: skips and exits 0 with no key; 11/11 with a key even under the leaked state. Re-introducing the #155 bug still fails 3 assertions, so the regression value is intact. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-08 23:06:52 -04:00
Dave HortonandClaude Opus 5	c066840f85	fix(inworld): read pitch and speakingRate from audioConfig in the say: params (#155 ) The streaming say: path guarded on opts.audioConfig?.pitch and opts.audioConfig?.speakingRate but interpolated opts.pitch and opts.speakingRate, which are undefined — so anyone setting them under audioConfig (what the docs and the portal defaults tell you to do) got 'pitch=undefined,speakingRate=undefined' on the wire and their setting silently dropped. Adds a test for the say: params that needs no credentials, since the streaming branch builds the path without calling the vendor. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-08 22:49:50 -04:00
Dave Horton	d89457f67d	1.0.15	2026-08-06 10:23:47 -04:00
Dave HortonandClaude Opus 5	3c15976669	fix(deps): remove junk "23" dependency (#154 ) "23": "^0.0.0" is not a real dependency - it is an empty placeholder package (0.0.0, no deps, unrelated third-party maintainer) that landed here from a stray npm install. Nothing in the package references it. Beyond the noise, it is a small supply-chain liability: a dependency on a squatted single-number name owned by nobody we know, shipped to every consumer of speech-utils. Lint passes; full test suite passes 105/105 with live vendor credentials. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-06 10:23:31 -04:00
Dave Horton	70abb04941	1.0.14	2026-08-06 10:13:51 -04:00
Dave HortonandClaude Opus 5	724af7fcab	fix(deps): drop unused undici dependency (#153 ) undici was declared as a direct dependency but is never required anywhere in this package - grep across index.js, lib/ and stubs/ finds no reference to undici, ProxyAgent, setGlobalDispatcher or Dispatcher. The one HTTP call in lib/synth-audio.js uses the global fetch, and Azure proxy support goes through the SDK's own setProxy plus the http_proxy_ip/http_proxy_port params. Removing it clears all seven open undici advisories from npm audit for this package and its consumers, and stops speech-utils pulling a duplicate undici 7.x into trees where the consumer already depends on undici 8 (feature-server does). Lint passes; full test suite passes 105/105 with live vendor credentials. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-06 10:13:31 -04:00
Dave Horton	c0a79090c2	1.0.13	2026-08-06 10:04:26 -04:00
Dave HortonandClaude Opus 5	7a9a3e2468	fix(deps): bump microsoft-cognitiveservices-speech-sdk to ^1.51.0 (#152 ) The Azure speech SDK was pinned to exactly 1.38.0 since the initial commit. It pulls in uuid <11.1.1, which carries GHSA-w5hq-g745-h8pq (missing buffer bounds check in v3/v5/v6 when buf is provided). The advisory covers Azure SDK 1.14.0-1.50.0; 1.51.0 clears it. Loosened the exact pin to a caret range so future patches come in without another PR. The SDK surface used here (SpeechConfig, SpeechSynthesizer, ResultReason, CancellationDetails, SpeechSynthesisOutputFormat) is unchanged in 1.51.0. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-06 10:03:59 -04:00
Dave Horton	ab2b3f7fb2	1.0.12	2026-08-04 09:04:50 -04:00
Dave HortonandClaude Opus 5	e848d0276e	feat(tts): add gradium as a TTS vendor (#151 ) Streaming arm returns a say: url for the mediajam dialect; the cache-render arm posts to /api/post/speech/tts with only_audio and pcm_8000, which is bare r8 samples and avoids gradium's streaming wav header (0xffffffff RIFF size). Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-04 09:02:07 -04:00
Dave Horton	d2447d477d	1.0.11	2026-08-03 12:04:47 -04:00
Dave HortonandClaude Opus 5	ec8bbacc2c	feat(tts): add nineninesix.ai synthesis (#150 ) Streaming goes through mediajam's say: url; the cache render posts to /tts/bytes for wav, since the service rejects mp3. Co-authored-by: Claude Opus 5 <noreply@anthropic.com>	2026-08-03 12:03:52 -04:00
Dave Horton	9694714a7f	1.0.10	2026-07-28 20:59:05 -04:00
Hoan Luu Huu	6a6de9b7a1	support google tts configuraiton (#149 )	2026-07-28 20:58:26 -04:00
Dave Horton	04d1fb548b	1.0.9	2026-07-21 06:52:05 -04:00
Hoan Luu Huu	bbf0167b40	support deepgramflux tts (#148 )	2026-07-21 06:49:38 -04:00
Dave Horton	4cfa286242	1.0.8	2026-07-05 17:42:09 -04:00
Hoan Luu Huu	42ac63adc6	support xaiTTS (#147 )	2026-07-05 17:41:36 -04:00
Dave Horton	12a121672d	1.0.7	2026-07-03 07:12:20 -04:00
Hoan Luu HuuandClaude Opus 4.8	7b94a5a969	feat(murf): add Murf.ai TTS support to synthAudio (#146 ) * feat(murf): add Murf.ai TTS support to synthAudio Add synthMurf() following the rimelabs/cartesia pattern: - streaming path returns a say:{vendor=murf,...} filePath consumed by the FreeSWITCH mod_murf_tts module - non-streaming path calls POST /v1/speech/stream (api-key header) and returns WAV audio for cache rendering Register murf in the supported-vendor assert list and the synth switch. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * fix(murf): drop Accept: audio/basic header (caused 406 Not Acceptable) Murf's /v1/speech/stream rejects an unmatched Accept header with 406; the response container is chosen by the `format` body field instead. Verified a WAV request now returns 200 (valid RIFF/WAVE). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-07-03 07:11:39 -04:00
Dave Horton	c47b4883c7	1.0.6	2026-06-17 16:20:30 -04:00
Dave HortonandClaude Opus 4.8	7d076bb8b4	chore: deprecate + remove verbio, nuance, playht speech vendor support (#144 ) * chore: deprecate and remove verbio, nuance speech vendor support Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> * chore: also deprecate and remove PlayHT speech vendor PlayHT was acquired and no longer provides the service. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	2026-06-17 16:20:00 -04:00
Dave HortonandClaude Opus 4.8	142323a151	fix(synth-audio): include the Polly VoiceId in the aws say: url synthPolly received voice but never emitted it into the say:{...} url (unlike synthMicrosoft, which includes voice=). Polly's SynthesizeSpeech requires a VoiceId, so the media-server-native (say:-url) Polly path had no way to pick a voice. Add voice= to the params. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>	2026-06-13 22:31:57 -04:00