support elevenlabs stt

lib/utils/speech-utils.js

@@ -9,6 +9,7 @@ const Houndify = require('houndify');
 const { GladiaClient } = require('@gladiaio/sdk');
 const {decrypt, obscureKey} = require('./encrypt-decrypt');
 const { RealtimeSession } = require('speechmatics');
+const {ElevenLabsClient} = require('@elevenlabs/elevenlabs-js');
 
 const TtsGoogleLanguagesVoices = require('./speech-data/tts-google');
 const TtsAwsLanguagesVoices = require('./speech-data/tts-aws');
@@ -52,11 +53,13 @@ const SttVoxistLanguagesVoices = require('./speech-data/stt-voxist');
 const SttVerbioLanguagesVoices = require('./speech-data/stt-verbio');
 const SttOpenaiLanguagesVoices = require('./speech-data/stt-openai');
 const SttGladiaLanguagesVoices = require('./speech-data/stt-gladia');
+const SttElevenlabsLanguagesVoices = require('./speech-data/stt-elevenlabs');
 
 
 const SttModelOpenai = require('./speech-data/stt-model-openai');
 const sttModelDeepgram = require('./speech-data/stt-model-deepgram');
 const sttModelCartesia = require('./speech-data/stt-model-cartesia');
+const sttModelElevenlabs = require('./speech-data/stt-model-elevenlabs');
 
 function capitalizeFirst(str) {
   if (!str) return str;
@@ -402,6 +405,42 @@ const testElevenlabs = async(logger, credentials) => {
   return response.body;
 };
 
+const testElevenlabsStt = async(logger, credentials) => {
+  const {api_key, stt_model_id} = credentials;
+
+  // Validate API key
+  if (!api_key) {
+    throw new Error('ElevenLabs API key is required');
+  }
+
+  const client = new ElevenLabsClient({
+    environment: 'https://api.elevenlabs.io',
+    apiKey: api_key
+  });
+
+  return new Promise(async(resolve, reject) => {
+    try {
+      const audioBuffer = fs.readFileSync(`${__dirname}/../../data/test_audio.wav`);
+
+      const result = await client.speechToText.convert({
+        file: audioBuffer,
+        modelId: stt_model_id
+      });
+
+      // Check if we got a valid response with text
+      if (result && result.text && result.text.length > 0) {
+        logger.debug({result}, 'ElevenLabs STT successful');
+        resolve(result);
+      } else {
+        reject(new Error('no transcript returned from ElevenLabs STT'));
+      }
+    } catch (error) {
+      logger.info({error}, 'failed to get ElevenLabs STT transcript');
+      reject(error);
+    }
+  });
+};
+
 const testPlayHT = async(logger, synthAudio, credentials) => {
   try {
     await synthAudio(
@@ -849,6 +888,7 @@ function decryptCredential(obj, credential, logger, isObscureKey = true) {
     obj.model_id = o.model_id;
     obj.api_uri = o.api_uri;
     obj.options = o.options;
+    obj.stt_model_id = o.stt_model_id;
   } else if ('playht' === obj.vendor) {
     const o = JSON.parse(decrypt(credential));
     obj.api_key = isObscureKey ? obscureKey(o.api_key) : o.api_key;
@@ -1218,13 +1258,13 @@ async function getLanguagesVoicesForElevenlabs(credential) {
         language.voices = voices;
       }
     }
-    return tranform(languages, undefined, models);
+    return tranform(languages, SttElevenlabsLanguagesVoices, models, sttModelElevenlabs);
   } else {
     const voices = TtsElevenlabsLanguagesVoices[0].voices;
     for (const language of TtsElevenlabsLanguagesVoices) {
       language.voices = voices;
     }
-    return tranform(TtsElevenlabsLanguagesVoices, undefined, TtsModelElevenLabs);
+    return tranform(TtsElevenlabsLanguagesVoices, SttElevenlabsLanguagesVoices, TtsModelElevenLabs, sttModelElevenlabs);
   }
 }
 
@@ -1783,6 +1823,7 @@ module.exports = {
   testIbmStt,
   testSonioxStt,
   testElevenlabs,
+  testElevenlabsStt,
   testPlayHT,
   testRimelabs,
   testInworld,