1.0.21

- streaming returns a say: url for the mediajam starter
- cache render POSTs /v1/audio/stream for raw 24 kHz PCM (r24)
- credential options merge under the verb's options
- no model: simba-3.2 for English, simba-3.0 otherwise

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -7,11 +7,18 @@ on:
 jobs:
   build:
     runs-on: ubuntu-latest
+    permissions:
+      id-token: write   # required to request the GitHub OIDC token for AWS
+      contents: read
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-node@v4
         with:
           node-version: '20'
+      - uses: aws-actions/configure-aws-credentials@v4
+        with:
+          role-to-assume: ${{ secrets.AWS_ROLE_ARN }}
+          aws-region: us-east-1
       - run: npm install
       - run: npm run jslint
       - run: sudo apt update && sudo apt install -y squid
@@ -19,10 +26,12 @@ jobs:
       - run: sudo systemctl start squid
       - run: npm test
         env:
-          AWS_ACCESS_KEY_ID: ${{ secrets.AWS_ACCESS_KEY_ID }}
-          AWS_REGION: ${{ secrets.AWS_REGION }}
-          AWS_SECRET_ACCESS_KEY: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
+          # AWS_* are exported by configure-aws-credentials above as short-lived
+          # OIDC credentials; no AWS keys are stored as repository secrets.
           GCP_JSON_KEY: ${{ secrets.GCP_JSON_KEY }}
+          # Enables the "AWS speech synth tests by RoleArn" test, which exercises the
+          # AssumeRole credential path used in production.
+          AWS_ROLE_ARN: ${{ secrets.AWS_ROLE_ARN }}
 
           MICROSOFT_API_KEY: ${{ secrets.MICROSOFT_API_KEY }}
           MICROSOFT_REGION: ${{ secrets.MICROSOFT_REGION }}

CLAUDE.md

@@ -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

lib/synth-audio.js

@@ -80,7 +80,8 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
   logger = logger || noopLogger;
 
   assert.ok(['google', 'aws', 'polly', 'microsoft', 'wellsaid', 'nvidia', 'elevenlabs',
-    'whisper', 'deepgram', 'deepgramflux', 'rimelabs', 'cartesia', 'inworld', 'resemble', 'murf', 'xai']
+    'whisper', 'deepgram', 'deepgramflux', 'rimelabs', 'cartesia', 'gradium', 'kugelaudio', 'nineninesix', 'inworld',
+    'resemble', 'murf', 'xai', 'fishaudio', 'speechify']
     .includes(vendor) ||
   vendor.startsWith('custom'),
   `synthAudio supported vendors are google, aws, microsoft, nvidia and wellsaid ..etc, not ${vendor}`);
@@ -135,12 +136,29 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
   } 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 ('speechify' === vendor) {
+    assert.ok(voice, 'synthAudio requires voice when speechify is used');
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when speechify is used');
+  } else if ('kugelaudio' === vendor) {
+    assert.ok(voice, 'synthAudio requires voice when kugelaudio is used');
+    assert.ok(credentials.api_key, 'synthAudio requires api_key when kugelaudio 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({
@@ -212,6 +230,31 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
           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 'kugelaudio':
+        audioData = await synthKugelaudio(logger, {
+          credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
+          disableTtsCache});
+        break;
+      case 'speechify':
+        audioData = await synthSpeechify(logger, {
+          credentials, options, stats, language, 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,
@@ -383,6 +426,32 @@ 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,
@@ -418,6 +487,15 @@ 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 {
@@ -486,6 +564,9 @@ 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 {
@@ -885,8 +966,9 @@ 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 {
@@ -1385,6 +1467,357 @@ 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;
+  }
+};
+
+/* kugelaudio — json websocket (/ws/tts/stream) for streaming, and POST /v1/tts/generate
+   for the cache render. the POST streams back bare little-endian 16-bit samples at the
+   requested sample_rate, which is exactly the r8 container at 8000.
+
+   voices are numeric ids (or public handles). language is an ISO 639-1 code that drives
+   text normalization; jambonz carries BCP-47, so only the primary subtag is sent. the
+   api rejects codes outside its list, so an unsupported or unset language is omitted
+   and the voice's own language applies. options.api_uri pins a region
+   (e.g. api.eu.kugelaudio.com).
+*/
+const KUGELAUDIO_LANGUAGES = ['ar', 'bg', 'bn', 'cs', 'da', 'de', 'el', 'en', 'es', 'fa', 'fi', 'fr', 'he', 'hi',
+  'hr', 'hu', 'id', 'it', 'ja', 'ko', 'ms', 'nl', 'no', 'pl', 'pt', 'ro', 'ru', 'sk', 'sl', 'sr', 'sv', 'ta', 'th',
+  'tr', 'uk', 'ur', 'vi', 'yue', 'zh'];
+const synthKugelaudio = async(logger, {
+  credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
+  disableTtsCache
+}) => {
+  const {api_key, model_id} = credentials;
+  const {api_uri, speed, cfg_scale, temperature, normalize, project_id, dictionary_ids} = options || {};
+  const isSet = (v) => v !== null && v !== undefined;
+
+  /* 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=kugelaudio';
+    params += `,voice=${voice}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    params += `,model_id=${model_id || 'kugel-3'}`;
+    if (language) params += `,language=${language}`;
+    if (api_uri) params += `,api_uri=${api_uri}`;
+    if (isSet(speed)) params += `,speed=${speed}`;
+    if (isSet(cfg_scale)) params += `,cfg_scale=${cfg_scale}`;
+    if (isSet(temperature)) params += `,temperature=${temperature}`;
+    if (isSet(normalize)) params += `,normalize=${normalize}`;
+    if (isSet(project_id)) params += `,project_id=${project_id}`;
+    /* the say: param parser is bracket-aware, so the json array survives intact */
+    if (Array.isArray(dictionary_ids)) params += `,dictionary_ids=${JSON.stringify(dictionary_ids)}`;
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+
+  try {
+    const sampleRate = 8000;
+    const host = (api_uri || 'api.kugelaudio.com').replace(/^[a-z]+:\/\//, '').replace(/\/$/, '');
+    const post = bent(`https://${host}`, 'POST', 'buffer', {
+      'Authorization': `Bearer ${api_key}`,
+      'Content-Type': 'application/json; charset=utf-8'
+    });
+    const voiceId = /^\d+$/.test(`${voice}`) ? Number(voice) : voice;
+    const lang = language && language.split('-')[0].toLowerCase();
+    const audioContent = await post('/v1/tts/generate', {
+      text,
+      voice_id: voiceId,
+      model_id: model_id || 'kugel-3',
+      sample_rate: sampleRate,
+      ...(KUGELAUDIO_LANGUAGES.includes(lang) && {language: lang}),
+      ...(isSet(speed) && {speed: Number(speed)}),
+      ...(isSet(cfg_scale) && {cfg_scale: Number(cfg_scale)}),
+      ...(isSet(temperature) && {temperature: Number(temperature)}),
+      ...(isSet(normalize) && {normalize: normalize === true || normalize === 'true'}),
+      ...(isSet(project_id) && {project_id: Number(project_id)}),
+      ...(Array.isArray(dictionary_ids) && {dictionary_ids})
+    });
+    return {
+      audioContent,
+      extension: 'r8',
+      sampleRate
+    };
+  } catch (err) {
+    logger.info({err}, 'synth kugelaudio returned error');
+    stats.increment('tts.count', ['vendor:kugelaudio', 'accepted:no']);
+    throw err;
+  }
+};
+
+/* simba-3.2 is English only; with no model set, other languages get simba-3.0 */
+const speechifyModel = (model_id, language) => {
+  if (model_id) return model_id;
+  return !language || /^en/i.test(language) ? 'simba-3.2' : 'simba-3.0';
+};
+const synthSpeechify = async(logger, {
+  credentials, options, stats, language, voice, key, text, renderForCaching, disableTtsStreaming,
+  disableTtsCache
+}) => {
+  const {api_key, model_id} = credentials;
+  let credOptions = credentials.options || {};
+  if (typeof credOptions === 'string') {
+    try {
+      credOptions = JSON.parse(credOptions);
+    } catch {
+      credOptions = {};
+    }
+  }
+  const {api_uri, loudness_normalization, text_normalization} = {...credOptions, ...options};
+  const isSet = (v) => v !== null && v !== undefined;
+  const model = speechifyModel(options?.model_id || model_id, language);
+
+  /* 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=${key}`;
+    params += ',vendor=speechify';
+    params += `,voice=${voice}`;
+    params += `,write_cache_file=${disableTtsCache ? 0 : 1}`;
+    params += `,model_id=${model}`;
+    if (language) params += `,language=${language}`;
+    if (api_uri) params += `,api_uri=${api_uri}`;
+    if (isSet(loudness_normalization)) params += `,loudness_normalization=${loudness_normalization}`;
+    if (isSet(text_normalization)) params += `,text_normalization=${text_normalization}`;
+    params += '}';
+
+    return {
+      filePath: `say:${params}${text.replace(/\n/g, ' ').replace(/\r/g, ' ')}`,
+      servedFromCache: false,
+      rtt: 0
+    };
+  }
+
+  try {
+    /* other pcm rates are mislabelled 24 kHz on workspaces pinned before 2026-09-30 */
+    const sampleRate = 24000;
+    const host = (api_uri || 'api.speechify.ai').replace(/^[a-z]+:\/\//, '').replace(/\/$/, '');
+    const post = bent(`https://${host}`, 'POST', 'buffer', {
+      'Authorization': `Bearer ${api_key}`,
+      'Content-Type': 'application/json',
+      'Speechify-Caller': 'jambonz'
+    });
+    const toBool = (v) => v === true || v === 'true';
+    const speechOptions = {
+      ...(isSet(loudness_normalization) && {loudness_normalization: toBool(loudness_normalization)}),
+      ...(isSet(text_normalization) && {text_normalization: toBool(text_normalization)})
+    };
+    const audioContent = await post('/v1/audio/stream', {
+      input: text,
+      voice_id: voice,
+      model,
+      output_format: `pcm_${sampleRate}`,
+      ...(language && {language}),
+      ...(Object.keys(speechOptions).length && {options: speechOptions})
+    });
+    return {
+      audioContent,
+      extension: 'r24',
+      sampleRate
+    };
+  } catch (err) {
+    logger.info({err}, 'synth speechify returned error');
+    stats.increment('tts.count', ['vendor:speechify', '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
 }) => {

package-lock.json

@@ -1,15 +1,14 @@
 {
   "name": "@jambonz/speech-utils",
-  "version": "1.0.9",
+  "version": "1.0.21",
   "lockfileVersion": 2,
   "requires": true,
   "packages": {
     "": {
       "name": "@jambonz/speech-utils",
-      "version": "1.0.9",
+      "version": "1.0.21",
       "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",
@@ -19,9 +18,8 @@
         "bent": "^7.3.12",
         "debug": "^4.3.4",
         "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",
@@ -760,6 +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",
@@ -2391,11 +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",
@@ -5278,17 +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": {
@@ -5312,36 +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": {
@@ -7031,15 +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",
@@ -7342,11 +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",
@@ -7924,6 +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",
@@ -9118,6 +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",
@@ -11106,16 +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": {
@@ -11135,15 +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=="
         }
       }
     },
@@ -12347,11 +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",

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@jambonz/speech-utils",
-  "version": "1.0.9",
+  "version": "1.0.21",
   "description": "TTS-related speech utilities for jambonz",
   "main": "index.js",
   "author": "Dave Horton",
@@ -25,7 +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",
@@ -35,9 +34,8 @@
     "bent": "^7.3.12",
     "debug": "^4.3.4",
     "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",

test/aws-credentials.js

@@ -0,0 +1,23 @@
+/**
+ * Resolve AWS credentials for the test suite.
+ *
+ * Returns null when AWS is not configured, so the caller skips.
+ *
+ * When AWS_SESSION_TOKEN is set the credentials are temporary -- GitHub OIDC in CI, or
+ * `aws sso login` locally -- and the key pair must not be passed through. Doing so sends
+ * lib/get-aws-sts-token.js down its accessKeyId branch, which calls GetSessionToken, and
+ * AWS rejects GetSessionToken when it is called with session credentials. Returning the
+ * region alone routes to the SDK's default credential provider chain instead, which
+ * handles temporary credentials correctly.
+ */
+module.exports = () => {
+  const region = process.env.AWS_REGION;
+  if (!region) return null;
+
+  const accessKeyId = process.env.AWS_ACCESS_KEY_ID;
+  const secretAccessKey = process.env.AWS_SECRET_ACCESS_KEY;
+
+  if (process.env.AWS_SESSION_TOKEN) return {region};
+  if (accessKeyId && secretAccessKey) return {accessKeyId, secretAccessKey, region};
+  return {region};
+};

test/aws.js

@@ -6,33 +6,33 @@ process.on('unhandledRejection', (reason, p) => {
   console.log('Unhandled Rejection at: Promise', p, 'reason:', reason);
 });
 
+const awsCredentials = require('./aws-credentials');
+
 const sleep = (ms) => new Promise((resolve) => setTimeout(resolve, ms));
 
 test('AWS - create and cache auth token', async(t) => {
   const fn = require('..');
   const {client, getAwsAuthToken} = fn(opts, logger);
 
-  if (!process.env.AWS_ACCESS_KEY_ID || !process.env.AWS_SECRET_ACCESS_KEY || !process.env.AWS_REGION) {
+  const credentials = awsCredentials();
+  if (!credentials) {
       t.pass('skipping AWS auth token tests since no AWS credentials provided');
       t.end();
       client.quit();
       return;
   }
+  // getAwsAuthToken derives its cache key from roleArn || accessKeyId || speech_credential_sid.
+  // With temporary credentials none of the first two are passed, so supply a stable sid --
+  // which is what production does for instance-profile credentials.
+  const args = {...credentials, speech_credential_sid: 'test-aws-speech-credential'};
+
   try {
-    let obj = await getAwsAuthToken({
-      accessKeyId: process.env.AWS_ACCESS_KEY_ID, 
-      secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
-      region: process.env.AWS_REGION
-    });
+    let obj = await getAwsAuthToken(args);
     //console.log({obj}, 'received auth token from AWS');
     t.ok(obj.securityToken && !obj.servedFromCache, 'successfullY generated auth token from AWS');
 
     await sleep(250);
-    obj = await getAwsAuthToken({
-      accessKeyId: process.env.AWS_ACCESS_KEY_ID, 
-      secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
-      region: process.env.AWS_REGION
-    });
+    obj = await getAwsAuthToken(args);
     //console.log({obj}, 'received auth token from AWS - second request');
     t.ok(obj.securityToken && obj.servedFromCache, 'successfully received access token from cache');
  

test/list-voices.js

@@ -1,4 +1,5 @@
 const test = require('tape').test ;
+const awsCredentials = require('./aws-credentials');
 const config = require('config');
 const opts = config.get('redis');
 const fs = require('fs');
@@ -45,18 +46,15 @@ test('AWS tests', async(t) => {
   const fn = require('..');
   const {client, getTtsVoices} = fn(opts, logger);
 
-  if (!process.env.AWS_ACCESS_KEY_ID || !process.env.AWS_SECRET_ACCESS_KEY || !process.env.AWS_REGION) {
-    t.pass('skipping AWS speech synth tests since AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, or AWS_REGION not provided');
+  const credentials = awsCredentials();
+  if (!credentials) {
+    t.pass('skipping AWS speech synth tests since AWS_REGION not provided');
     return t.end();
   }
   try {
     const opts = {
       vendor: 'aws',
-      credentials: {
-        accessKeyId: process.env.AWS_ACCESS_KEY_ID,
-        secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
-        region: process.env.AWS_REGION,
-      }
+      credentials
     };
     let result = await getTtsVoices(opts);
     t.ok(result?.Voices?.length > 0, `GetVoices: successfully retrieved ${result.Voices.length} voices from AWS`);

test/synth.js

@@ -1,4 +1,5 @@
 const test = require('tape').test;
+const awsCredentials = require('./aws-credentials');
 const config = require('config');
 const opts = config.get('redis');
 const fs = require('fs');
@@ -442,6 +443,95 @@ 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 +616,42 @@ 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);
@@ -543,18 +669,15 @@ test('AWS speech synth tests', async(t) => {
   const fn = require('..');
   const {synthAudio, client} = fn(opts, logger);
 
-  if (!process.env.AWS_ACCESS_KEY_ID || !process.env.AWS_SECRET_ACCESS_KEY || !process.env.AWS_REGION) {
-    t.pass('skipping AWS speech synth tests since AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, or AWS_REGION not provided');
+  const credentials = awsCredentials();
+  if (!credentials) {
+    t.pass('skipping AWS speech synth tests since AWS_REGION not provided');
     return t.end();
   }
   try {
     let opts = await synthAudio(stats, {
       vendor: 'aws',
-      credentials: {
-        accessKeyId: process.env.AWS_ACCESS_KEY_ID,
-        secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
-        region: process.env.AWS_REGION,
-      },
+      credentials,
       language: 'en-US',
       voice: 'Joey',
       text: 'This is a test.  This is only a test',
@@ -564,11 +687,7 @@ test('AWS speech synth tests', async(t) => {
 
     opts = await synthAudio(stats, {
       vendor: 'aws',
-      credentials: {
-        accessKeyId: process.env.AWS_ACCESS_KEY_ID,
-        secretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
-        region: process.env.AWS_REGION,
-      },
+      credentials,
       language: 'en-US',
       voice: 'Joey',
       text: 'This is a test.  This is only a test',
@@ -599,9 +718,18 @@ test('AWS speech synth tests by RoleArn', async(t) => {
       },
       language: 'en-US',
       voice: 'Joey',
-      text: 'This is a test.  This is only a test',
+      // Distinct text on purpose: the 'AWS speech synth tests' above cache audio for
+      // the same vendor/voice/language, and identical text would hit that cache entry,
+      // making servedFromCache true and this assertion fail.
+      text: 'This is a roleArn test.  This is only a roleArn test',
+      // Without this, synthPolly returns the mediajam streaming params string, which
+      // embeds accessKeyId/secretAccessKey/sessionToken (see lib/synth-audio.js).
+      // Every other synth test in this file renders for caching; this one did not,
+      // so it printed live credentials into a public CI log.
+      renderForCaching: true,
     });
-    t.ok(!opts.servedFromCache, `successfully synthesized aws by roleArn audio to ${opts.filePath}`);
+    // Deliberately does not interpolate opts.filePath -- it can carry credentials.
+    t.ok(!opts.servedFromCache, 'successfully synthesized aws by roleArn audio');
   } catch (err) {
     console.error(err);
     t.end(err);
@@ -933,6 +1061,165 @@ 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('kugelaudio speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.KUGELAUDIO_API_KEY) {
+    t.pass('skipping kugelaudio speech synth tests since KUGELAUDIO_API_KEY is not provided');
+    return t.end();
+  }
+  const text = 'Guten Tag und willkommen bei jambonz! Ihre Bestellung kostet 12,99 Euro. ' + Date.now();
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'kugelaudio',
+      credentials: {
+        api_key: process.env.KUGELAUDIO_API_KEY,
+        model_id: 'kugel-3'
+      },
+      language: 'de-DE',
+      voice: '1930',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache, `successfully synthed kugelaudio audio to ${opts.filePath}`);
+
+  } catch (err) {
+    console.error(JSON.stringify(err));
+    t.end(err);
+  }
+  client.quit();
+});
+
+test('speechify speech synth tests', async(t) => {
+  const fn = require('..');
+  const {synthAudio, client} = fn(opts, logger);
+
+  if (!process.env.SPEECHIFY_API_KEY) {
+    t.pass('skipping speechify speech synth tests since SPEECHIFY_API_KEY is not provided');
+    return t.end();
+  }
+  const text = 'Hi there and welcome to jambonz! ' + Date.now();
+  try {
+    const opts = await synthAudio(stats, {
+      vendor: 'speechify',
+      credentials: {
+        api_key: process.env.SPEECHIFY_API_KEY
+      },
+      language: 'en-US',
+      voice: 'geffen_32',
+      text,
+      renderForCaching: true
+    });
+    t.ok(!opts.servedFromCache, `successfully synthed speechify 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);
@@ -963,6 +1250,69 @@ 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);