support mod cartesia transcribe (#463)

lib/routes/api/speech-credentials.js

@@ -135,6 +135,8 @@ const encryptCredential = (obj) => {
     custom_tts_endpoint,
     custom_tts_endpoint_url,
     use_custom_stt,
+    use_for_stt,
+    use_for_tts,
     custom_stt_endpoint,
     custom_stt_endpoint_url,
     tts_api_key,
@@ -148,7 +150,10 @@ const encryptCredential = (obj) => {
     custom_tts_streaming_url,
     auth_token = '',
     cobalt_server_uri,
+    // For most vendors, model_id is being used for both TTS and STT, or one of them.
+    // for Cartesia, model_id is used for TTS only. introduce stt_model_id for STT
     model_id,
+    stt_model_id,
     user_id,
     voice_engine,
     engine_version,
@@ -259,8 +264,17 @@ const encryptCredential = (obj) => {
 
     case 'cartesia':
       assert(api_key, 'invalid cartesia speech credential: api_key is required');
+      if (use_for_tts) {
         assert(model_id, 'invalid cartesia speech credential: model_id is required');
-      const cartesiaData = JSON.stringify({api_key, model_id, options});
+      }
+      if (use_for_stt) {
+        assert(stt_model_id, 'invalid cartesia speech credential: stt_model_id is required');
+      }
+      const cartesiaData = JSON.stringify({
+        api_key,
+        ...(model_id && {model_id}),
+        ...(stt_model_id && {stt_model_id}),
+        options});
       return encrypt(cartesiaData);
 
     case 'rimelabs':
@@ -487,6 +501,7 @@ router.put('/:sid', async(req, res) => {
           custom_tts_streaming_url,
           cobalt_server_uri,
           model_id,
+          stt_model_id,
           voice_engine,
           options,
           deepgram_stt_uri,
@@ -518,6 +533,7 @@ router.put('/:sid', async(req, res) => {
           custom_tts_streaming_url,
           cobalt_server_uri,
           model_id,
+          stt_model_id,
           voice_engine,
           options,
           deepgram_stt_uri,
@@ -833,17 +849,28 @@ router.get('/:sid/test', async(req, res) => {
         }
       }
     } else if (cred.vendor === 'cartesia') {
-      if (cred.use_for_tts) {
+      if (cred.use_for_tts || cred.use_for_stt) {
         try {
+          // Cartesia does not have API for testing STT, same key is used for both TTS and STT
           await testCartesia(logger, synthAudio, credential);
+          if (cred.use_for_tts) {
             results.tts.status = 'ok';
+          }
+          if (cred.use_for_stt) {
+            results.stt.status = 'ok';
+          }
           SpeechCredential.ttsTestResult(sid, true);
         } catch (err) {
           let reason = err.message;
           try {
             reason = await err.text();
           } catch {}
+          if (cred.use_for_tts) {
             results.tts = {status: 'fail', reason};
+          }
+          if (cred.use_for_stt) {
+            results.stt = {status: 'fail', reason};
+          }
           SpeechCredential.ttsTestResult(sid, false);
         }
       }

lib/utils/speech-data/stt-model-cartesia.js

@@ -0,0 +1,4 @@
+module.exports = [
+  { name: 'Ink-whisper', value: 'ink-whisper' },
+];
+

lib/utils/speech-utils.js

@@ -49,6 +49,7 @@ const SttOpenaiLanguagesVoices = require('./speech-data/stt-openai');
 
 const SttModelOpenai = require('./speech-data/stt-model-openai');
 const sttModelDeepgram = require('./speech-data/stt-model-deepgram');
+const sttModelCartesia = require('./speech-data/stt-model-cartesia');
 
 function capitalizeFirst(str) {
   if (!str) return str;
@@ -680,6 +681,7 @@ function decryptCredential(obj, credential, logger, isObscureKey = true) {
     const o = JSON.parse(decrypt(credential));
     obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
     obj.model_id = o.model_id;
+    obj.stt_model_id = o.stt_model_id;
     obj.options = o.options;
   } else if ('rimelabs' === obj.vendor) {
     const o = JSON.parse(decrypt(credential));
@@ -1403,9 +1405,23 @@ async function getLanguagesVoicesForCartesia(credential) {
       return acc;
     }, []);
 
-    return tranform(ttsVoices, undefined, TtsModelCartesia);
+    return tranform(
+      ttsVoices,
+      ttsVoices.map((voice) => ({
+        name: voice.name,
+        value: voice.value,
+      })),
+      TtsModelCartesia,
+      sttModelCartesia);
   }
-  return tranform(ttsCartesia, undefined, TtsModelCartesia);
+  return tranform(
+    ttsCartesia,
+    ttsCartesia.map((voice) => ({
+      name: voice.name,
+      value: voice.value,
+    })),
+    TtsModelCartesia,
+    sttModelCartesia);
 }
 
 module.exports = {