From 0bd3ff787bf084aa105e94c8ea65b5bc6eae782a Mon Sep 17 00:00:00 2001
From: Daniel Onyejesi <danielonyejesi@gmail.com>
Date: Sun, 29 Mar 2026 22:28:32 -0400
Subject: [PATCH] Fix STT/TTS model paths: use exact vertex_ai/ paths for
 LiteLLM routing
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

LiteLLM aliases (whisper-1, tts-1) don't resolve in audio endpoints —
only chat completions support alias routing. Use exact paths:
- STT default: vertex_ai/chirp (was whisper-1)
- TTS default: vertex_ai/text-to-speech (was tts-1)
Override via LITELLM_STT_MODEL / LITELLM_TTS_MODEL in .env.
---
 .env.example             | 6 +++---
 src/routes/transcribe.js | 2 +-
 src/routes/tts.js        | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/.env.example b/.env.example
index ab45597..74ede76 100644
--- a/.env.example
+++ b/.env.example
@@ -35,12 +35,12 @@ OPENROUTER_API_KEY=sk-or-v1-your-key
 #
 # LiteLLM Speech-to-Text (auto-enabled when LITELLM_API_BASE is set)
 # TRANSCRIBE_PROVIDER=litellm
-# LITELLM_STT_MODEL=whisper-1          # or: groq/whisper-large-v3, openai/whisper-1
+# LITELLM_STT_MODEL=vertex_ai/chirp    # or: vertex_ai/chirp_2 for newer model
 #
 # LiteLLM Text-to-Speech (auto-enabled when LITELLM_API_BASE is set)
 # TTS_PROVIDER=litellm
-# LITELLM_TTS_MODEL=tts-1              # or: tts-1-hd, openai/tts-1
-# LITELLM_TTS_VOICE=alloy              # alloy, echo, fable, onyx, nova, shimmer
+# LITELLM_TTS_MODEL=vertex_ai/text-to-speech   # voice is set inside LiteLLM per model
+# LITELLM_TTS_VOICE is ignored for Vertex AI — voice is configured in LiteLLM model entry
 
 # ============================================================
 # TRANSCRIPTION (speech-to-text)
diff --git a/src/routes/transcribe.js b/src/routes/transcribe.js
index 1f09980..d026387 100644
--- a/src/routes/transcribe.js
+++ b/src/routes/transcribe.js
@@ -70,7 +70,7 @@ router.post('/transcribe', authMiddleware, upload.single('audio'), async (req, r
 
     if (provider === 'litellm') {
       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 sttModel = process.env.LITELLM_STT_MODEL || 'vertex_ai/chirp';
       var FormData = require('form-data');
       var axios = require('axios');
       var form = new FormData();
diff --git a/src/routes/tts.js b/src/routes/tts.js
index fe00a9c..89e8e2e 100644
--- a/src/routes/tts.js
+++ b/src/routes/tts.js
@@ -26,7 +26,7 @@ router.post('/text-to-speech', authMiddleware, async (req, res) => {
 
     if (ttsProvider === 'litellm') {
       if (!process.env.LITELLM_API_BASE) return res.status(400).json({ error: 'LITELLM_API_BASE not set.' });
-      var ttsModel = process.env.LITELLM_TTS_MODEL || 'tts-1';
+      var ttsModel = process.env.LITELLM_TTS_MODEL || 'vertex_ai/text-to-speech';
       var base = process.env.LITELLM_API_BASE.replace(/\/+$/, '');
       var ttsHeaders = { 'Content-Type': 'application/json' };
       if (process.env.LITELLM_API_KEY) ttsHeaders['Authorization'] = 'Bearer ' + process.env.LITELLM_API_KEY;