From ce7d0e749dc5e98673835745146d6d3ba7bfafe5 Mon Sep 17 00:00:00 2001
From: Daniel Onyejesi <danielonyejesi@gmail.com>
Date: Sun, 29 Mar 2026 21:32:30 -0400
Subject: [PATCH] Fix LiteLLM STT: use axios directly instead of OpenAI SDK
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

OpenAI SDK's audio.transcriptions.create() hangs with LiteLLM
(no timeout, SDK-level incompatibility with multipart handling).
Use axios + form-data directly with 120s timeout — same approach
as ElevenLabs TTS. Handles both {text:"..."} and plain string responses.
---
 src/routes/transcribe.js | 28 +++++++++++++++++++---------
 1 file changed, 19 insertions(+), 9 deletions(-)

diff --git a/src/routes/transcribe.js b/src/routes/transcribe.js
index fcb6d73..e0706b9 100644
--- a/src/routes/transcribe.js
+++ b/src/routes/transcribe.js
@@ -69,17 +69,27 @@ router.post('/transcribe', authMiddleware, upload.single('audio'), async (req, r
     }
 
     if (provider === 'litellm') {
-      if (!litellmClient) return res.status(400).json({ error: 'LiteLLM not configured. Set LITELLM_API_BASE.' });
+      if (!process.env.LITELLM_API_BASE) return res.status(400).json({ error: 'LITELLM_API_BASE not set.' });
       var sttModel = process.env.LITELLM_STT_MODEL || 'whisper-1';
-      var file = new File([req.file.buffer], 'audio.webm', { type: req.file.mimetype || 'audio/webm' });
-      var result = await litellmClient.audio.transcriptions.create({
-        file: file,
-        model: sttModel,
-        language: 'en',
-        response_format: 'text',
-        prompt: 'Medical patient encounter. Pediatric. Clinical terms, diagnoses, medications.'
+      var FormData = require('form-data');
+      var axios = require('axios');
+      var form = new FormData();
+      form.append('file', req.file.buffer, {
+        filename: 'audio.webm',
+        contentType: req.file.mimetype || 'audio/webm'
       });
-      var text = typeof result === 'string' ? result : result.text;
+      form.append('model', sttModel);
+      form.append('language', 'en');
+      form.append('response_format', 'json');
+      form.append('prompt', 'Medical patient encounter. Pediatric. Clinical terms, diagnoses, medications.');
+      var sttHeaders = Object.assign({}, form.getHeaders());
+      if (process.env.LITELLM_API_KEY) sttHeaders['Authorization'] = 'Bearer ' + process.env.LITELLM_API_KEY;
+      var base = process.env.LITELLM_API_BASE.replace(/\/+$/, '');
+      var sttResp = await axios.post(base + '/v1/audio/transcriptions', form, {
+        headers: sttHeaders,
+        timeout: 120000
+      });
+      var text = (sttResp.data && sttResp.data.text) ? sttResp.data.text : String(sttResp.data || '');
       console.log('[Transcribe] LiteLLM (' + sttModel + ') done in ' + (Date.now() - startTime) + 'ms');
       return res.json({ success: true, text: text, provider: 'litellm-stt', duration: Date.now() - startTime });
     }