From a72002a84ee139db2bf82bd23a0d9ed6be7aaada Mon Sep 17 00:00:00 2001
From: Emerson Gomes <emerson.gomes@thalesgroup.com>
Date: Wed, 23 Sep 2026 04:40:54 -0500
Subject: [PATCH] fix(vertex-ai): use provider speech media type in proxy

---
 litellm/proxy/proxy_server.py | 16 +---------------
 1 file changed, 1 insertion(+), 15 deletions(-)

diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index ab66191ae25..6f67845f537 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -12150,24 +12150,10 @@ async def audio_speech(
         upstream_content_type: Final = (
             response.response.headers.get("content-type") if isinstance(response, HttpxBinaryResponseContent) else None
         )
-        resolved_media_type: Final = resolve_speech_media_type(
+        media_type: Final = extra_media_type or resolve_speech_media_type(
             upstream_content_type=upstream_content_type,
             response_format=requested_format_str,
         )
-        request_model: Final = data.get("model", "")
-        media_type: Final = (
-            extra_media_type
-            if extra_media_type is not None
-            else (
-                "audio/wav"
-                if (
-                    requested_format_str is None
-                    and isinstance(request_model, str)
-                    and litellm.utils.is_gemini_tts_model(request_model)
-                )
-                else resolved_media_type
-            )
-        )
 
         return StreamingResponse(
             _audio_speech_chunk_generator(response),