feat(tts): add speechify as a TTS vendor (#165)

- 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>

lib/synth-audio.js

@@ -81,7 +81,7 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
 
   assert.ok(['google', 'aws', 'polly', 'microsoft', 'wellsaid', 'nvidia', 'elevenlabs',
     'whisper', 'deepgram', 'deepgramflux', 'rimelabs', 'cartesia', 'gradium', 'kugelaudio', 'nineninesix', 'inworld',
-    'resemble', 'murf', 'xai', 'fishaudio']
+    'resemble', 'murf', 'xai', 'fishaudio', 'speechify']
     .includes(vendor) ||
   vendor.startsWith('custom'),
   `synthAudio supported vendors are google, aws, microsoft, nvidia and wellsaid ..etc, not ${vendor}`);
@@ -139,6 +139,9 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
   } 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');
@@ -242,6 +245,11 @@ async function synthAudio(client, createHash, retrieveHash, logger, stats, { acc
           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,
@@ -1602,6 +1610,84 @@ const synthKugelaudio = async(logger, {
   }
 };
 
+/* 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