fix litellm speech model discovery

docker-compose.yml

@@ -6,10 +6,15 @@ services:
       - "127.0.0.1:3552:3000"
     env_file:
       - .env
+      - /root/litellm/.env
     environment:
       CLINICAL_ASSISTANT_MCP_URL: http://mcp:8000/mcp
       REDIS_URL: redis://ped-ai-redis:6379
       LOKI_URL: http://monitoring-loki:3100
+      LITELLM_API_BASE: http://litellm:4000
+      TTS_PROVIDER: litellm
+      LITELLM_TTS_MODEL: local-kokoro-tts
+      LITELLM_TTS_VOICE: sherpa/kokoro:am_adam
     volumes:
       - scribe-logs:/app/data/logs
       - clinical-assistant-mcp-data:/app/mcp-data:ro
@@ -24,6 +29,7 @@ services:
       - default
       - mcp-server_default
       - monitoring_default
+      - speech_net
     healthcheck:
       test: ["CMD", "wget", "--spider", "-q", "http://localhost:3000/api/health"]
       interval: 30s
@@ -80,3 +86,5 @@ networks:
     external: true
   monitoring_default:
     external: true
+  speech_net:
+    external: true

src/routes/adminConfig.js

@@ -11,7 +11,7 @@ var logger = require('../utils/logger');
 var { gatewayUrl } = require('../utils/errors');
 var { GOOGLE_TTS_VOICES, OPENAI_TTS_VOICES, getTTSEnvProvider, getLiteLLMTTSModels, getTTSProvider, getTTSVoiceLists } = require('../utils/ttsProvider');
 var { getLiteLLMHeaders } = require('../utils/litellm');
-var { getSTTDependencies, getSTTModelLists, getSTTProvider } = require('../utils/sttProvider');
+var { getSTTDependencies, getLiteLLMSTTModels, getSTTModelLists, getSTTProvider } = require('../utils/sttProvider');
 
 router.use(authMiddleware);
 
@@ -478,9 +478,9 @@ router.get('/config/tts', async function(req, res) {
       envVoice: envVoice,
       envModel: envModel,
       configured: {
-        google: isGoogleTTSConfigured(),
+        google: false,
         litellm: !!process.env.LITELLM_API_BASE,
-        elevenlabs: !!process.env.ELEVENLABS_API_KEY
+        elevenlabs: false
       },
       voices: getTTSVoiceLists()
     });
@@ -519,32 +519,13 @@ router.get('/config/tts/discover', async function(req, res) {
     }
 
     if (provider === 'litellm' && process.env.LITELLM_API_BASE) {
-      // base URL handled by gatewayUrl() helper
       try {
-        var lResp = await axios.get(gatewayUrl('/models'), { headers: getLiteLLMHeaders(), timeout: 10000 });
+        var liteLLMBase = (process.env.LITELLM_API_BASE || '').replace(/\/+$/, '').replace(/\/v1\/?$/, '');
+        var lResp = await axios.get(liteLLMBase + '/model/info', { headers: getLiteLLMHeaders(), timeout: 10000 });
         getLiteLLMTTSModels(lResp.data && lResp.data.data).forEach(function(id) {
           discovered.push({ id: id, name: id, source: 'gateway-api' });
         });
       } catch (e) { logger.warn('LiteLLM TTS model list failed: ' + e.message); }
-
-      // OpenAI voices
-      OPENAI_TTS_VOICES.forEach(function(v) {
-        discovered.push({ id: v, name: v + ' (OpenAI voice)', source: 'openai-voices' });
-      });
-
-      // ElevenLabs voices — fetch from API if key available
-      if (process.env.ELEVENLABS_API_KEY) {
-        try {
-          var elResp = await axios.get('https://api.elevenlabs.io/v1/voices', {
-            headers: { 'xi-api-key': process.env.ELEVENLABS_API_KEY }, timeout: 10000
-          });
-          if (elResp.data && elResp.data.voices) {
-            elResp.data.voices.forEach(function(v) {
-              discovered.push({ id: v.voice_id, name: v.name + ' (ElevenLabs)', source: 'elevenlabs-api' });
-            });
-          }
-        } catch (e) { logger.warn('ElevenLabs voice fetch failed: ' + e.message); }
-      }
     }
 
     if (provider === 'elevenlabs' && process.env.ELEVENLABS_API_KEY) {
@@ -680,16 +661,17 @@ router.get('/config/stt/discover', async function(req, res) {
     if (provider === 'litellm' && process.env.LITELLM_API_BASE) {
       // base URL handled by gatewayUrl() helper
       try {
-        var sResp = await axios.get(gatewayUrl('/models'), { headers: getLiteLLMHeaders(), timeout: 10000 });
+        var liteLLMBase = (process.env.LITELLM_API_BASE || '').replace(/\/+$/, '').replace(/\/v1\/?$/, '');
-        if (sResp.data && sResp.data.data) {
+        var sResp = await axios.get(liteLLMBase + '/model/info', { headers: getLiteLLMHeaders(), timeout: 10000 });
-          sResp.data.data.forEach(function(m) {
+        getLiteLLMSTTModels(sResp.data && sResp.data.data).forEach(function(id) {
-            var mid = (m.id || '').toLowerCase();
+          discovered.push({ id: id, name: id, source: 'gateway-api' });
-            if (mid.indexOf('whisper') !== -1 || mid.indexOf('parakeet') !== -1 || mid.indexOf('gemini') !== -1 || mid.indexOf('audio') !== -1 || mid.indexOf('transcri') !== -1 || mid.indexOf('scribe') !== -1 || mid.indexOf('nova') !== -1 || mid.indexOf('deepgram') !== -1 || mid.indexOf('eleven') !== -1 || mid.indexOf('chirp') !== -1) {
+        });
-              discovered.push({ id: m.id, name: m.id, source: 'gateway-api' });
-            }
-          });
-        }
       } catch(e) { logger.warn('STT discovery failed: ' + e.message); }
+      if (discovered.length === 0) {
+        getSTTModelLists().litellm.forEach(function(id) {
+          discovered.push({ id: id, name: id, source: 'configured-default' });
+        });
+      }
     }
 
     if (provider === 'local') {

src/routes/tts.js

@@ -5,10 +5,11 @@ const { synthesizeWithGoogleTTS } = require('../utils/ttsGoogle');
 const { authMiddleware } = require('../middleware/auth');
 var logger = require('../utils/logger');
 var { gatewayUrl } = require('../utils/errors');
-var { getLiteLLMHeaders, getTTSProvider } = require('../utils/ttsProvider');
+var { getTTSProvider } = require('../utils/ttsProvider');
+var { getLiteLLMHeaders } = require('../utils/litellm');
 
 // Provider priority (auto-detect):
-// TTS_PROVIDER=litellm    → LiteLLM proxy at /v1/audio/speech (default when LITELLM_API_BASE is set)
+// TTS_PROVIDER=litellm    → LiteLLM proxy at /v1/audio/speech
 // TTS_PROVIDER=google     → Google Cloud TTS direct (HIPAA eligible, opt-in)
 // TTS_PROVIDER=elevenlabs → ElevenLabs direct (not HIPAA, opt-in)
 // Auto: litellm > google > elevenlabs

src/utils/litellm.js

@@ -1,7 +1,8 @@
 function getLiteLLMHeaders(contentType) {
   var headers = {};
   if (contentType) headers['Content-Type'] = contentType;
-  if (process.env.LITELLM_API_KEY) headers.Authorization = 'Bearer ' + process.env.LITELLM_API_KEY;
+  var key = process.env.LITELLM_API_KEY || process.env.LITELLM_MASTER_KEY;
+  if (key) headers.Authorization = 'Bearer ' + key;
   return headers;
 }
 

src/utils/sttProvider.js

@@ -41,6 +41,17 @@ function getSTTModelLists() {
   };
 }
 
+function isLiteLLMSTTModel(model) {
+  var mode = model && model.model_info && model.model_info.mode ? String(model.model_info.mode) : '';
+  return mode === 'audio_transcription';
+}
+
+function getLiteLLMSTTModels(models) {
+  return (models || [])
+    .filter(isLiteLLMSTTModel)
+    .map(function(model) { return model && (model.id || model.model_name) ? (model.id || model.model_name) : String(model || ''); });
+}
+
 module.exports = {
   AWS_STT_MODES,
   GOOGLE_STT_MODELS,
@@ -48,6 +59,8 @@ module.exports = {
   LOCAL_WHISPER_MODELS,
   OPENAI_STT_MODELS,
   getSTTDependencies,
+  getLiteLLMSTTModels,
   getSTTModelLists,
-  getSTTProvider
+  getSTTProvider,
+  isLiteLLMSTTModel
 };

src/utils/ttsProvider.js

@@ -32,10 +32,15 @@ function isOpenAITTSModel(modelId) {
   return /^openai[-/](tts-1|tts-1-hd|gpt-4o-mini-tts|gpt-audio)/.test(String(modelId || '').toLowerCase());
 }
 
+function isLiteLLMTTSModel(model) {
+  var mode = model && model.model_info && model.model_info.mode ? String(model.model_info.mode) : '';
+  return mode === 'audio_speech';
+}
+
 function getLiteLLMTTSModels(models) {
   return (models || [])
-    .map(function(model) { return model && model.id ? model.id : ''; })
+    .filter(isLiteLLMTTSModel)
-    .filter(isOpenAITTSModel);
+    .map(function(model) { return model && (model.id || model.model_name) ? (model.id || model.model_name) : String(model || ''); });
 }
 
 module.exports = {
@@ -47,5 +52,6 @@ module.exports = {
   getLiteLLMTTSModels,
   getTTSProvider,
   getTTSVoiceLists,
+  isLiteLLMTTSModel,
   isOpenAITTSModel
 };

test/stt-provider.test.js

@@ -46,3 +46,31 @@ test('STT model lists preserve admin defaults', () => {
     local: ['tiny', 'base', 'small', 'medium', 'large']
   });
 });
+
+test('LiteLLM STT discovery prefers audio_transcription metadata', () => {
+  const sttProvider = require('../src/utils/sttProvider');
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'custom-asr', model_info: { mode: 'audio_transcription' } }), true);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'local-parakeet-v3', model_info: { mode: 'audio_transcription' } }), true);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'local-chatterbox-turbo', model_info: { mode: 'audio_speech' } }), false);
+});
+
+test('LiteLLM STT discovery ignores names when metadata is absent', () => {
+  const sttProvider = require('../src/utils/sttProvider');
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'groq-whisper-large-v3' }), false);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'mistral-voxtral-mini-transcribe' }), false);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'openrouter-nova-pro-v1' }), false);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'openrouter-gemini-2.5-flash' }), false);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'openrouter-gpt-audio' }), false);
+  assert.equal(sttProvider.isLiteLLMSTTModel({ id: 'local-chatterbox-turbo' }), false);
+});
+
+test('LiteLLM STT model extraction filters by metadata only', () => {
+  const sttProvider = require('../src/utils/sttProvider');
+  assert.deepEqual(sttProvider.getLiteLLMSTTModels([
+    { model_name: 'custom-asr', model_info: { mode: 'audio_transcription' } },
+    { model_name: 'local-parakeet-v3', model_info: { mode: 'audio_transcription' } },
+    { model_name: 'local-chatterbox-turbo', model_info: { mode: 'audio_speech' } },
+    { id: 'groq-whisper-large-v3' },
+    { id: 'gpt-4.1' }
+  ]), ['custom-asr', 'local-parakeet-v3']);
+});

test/tts-provider.test.js

@@ -57,15 +57,28 @@ test('TTS model discovery keeps OpenAI speech models only', () => {
   assert.equal(ttsProvider.isOpenAITTSModel('vertex-gemini-2.5-flash-tts'), false);
 });
 
+test('TTS model discovery uses audio_speech metadata only', () => {
+  const ttsProvider = require('../src/utils/ttsProvider');
+  assert.equal(ttsProvider.isLiteLLMTTSModel({ id: 'local-chatterbox-turbo' }), false);
+  assert.equal(ttsProvider.isLiteLLMTTSModel({ id: 'local-kokoro-tts' }), false);
+  assert.equal(ttsProvider.isLiteLLMTTSModel({ model_name: 'local-chatterbox-turbo', model_info: { mode: 'audio_speech' } }), true);
+  assert.equal(ttsProvider.isLiteLLMTTSModel({ id: 'custom-provider-model', model_info: { mode: 'audio_speech' } }), true);
+  assert.equal(ttsProvider.isLiteLLMTTSModel({ id: 'local-parakeet-v3', model_info: { mode: 'audio_transcription' } }), false);
+});
+
 test('LiteLLM TTS model extraction filters gateway model objects', () => {
   const ttsProvider = require('../src/utils/ttsProvider');
   assert.deepEqual(ttsProvider.getLiteLLMTTSModels([
     { id: 'openai-tts-1' },
     { id: 'mistral-voxtral-mini-tts' },
     { id: 'openai-gpt-4o-mini-tts' },
+    { id: 'local-chatterbox-turbo' },
+    { model_name: 'local-chatterbox-turbo', model_info: { mode: 'audio_speech' } },
+    { id: 'custom-provider-model', model_info: { mode: 'audio_speech' } },
+    { id: 'local-parakeet-v3', model_info: { mode: 'audio_transcription' } },
     { id: 'vertex-gemini-2.5-flash-tts' },
     { id: 'gpt-4.1' }
-  ]), ['openai-tts-1', 'openai-gpt-4o-mini-tts']);
+  ]), ['local-chatterbox-turbo', 'custom-provider-model']);
 });
 
 test('LiteLLM headers include optional content type without exposing key value', () => {