From b67b44bdaaa747d6d40b24c2b9583b9caf332a38 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Sat, 29 Aug 2026 21:10:14 -0700
Subject: [PATCH] fix(proxy): map audio_speech errors to their status codes
 instead of a blanket 500

---
 litellm/proxy/proxy_server.py                 | 10 ++++++-
 .../proxy/proxy_server/test_routes_audio.py   | 30 +++++++++++++++++++
 2 files changed, 39 insertions(+), 1 deletion(-)

diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index ee1eb876b87..37034ea9a62 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -10996,7 +10996,15 @@ async def audio_speech(
         )
         verbose_proxy_logger.error("litellm.proxy.proxy_server.audio_speech(): Exception occured - %s", e)
         verbose_proxy_logger.debug(traceback.format_exc())
-        raise e
+        if isinstance(e, (ProxyException, HTTPException)):
+            raise e
+        raise ProxyException(
+            message=getattr(e, "message", f"{e}"),
+            type=getattr(e, "type", "None"),
+            param=getattr(e, "param", "None"),
+            openai_code=getattr(e, "code", None),
+            code=getattr(e, "status_code", 500),
+        )
 
 
 @router.post(
diff --git a/tests/test_litellm/proxy/proxy_server/test_routes_audio.py b/tests/test_litellm/proxy/proxy_server/test_routes_audio.py
index b99affc2ac3..522a20cd34b 100644
--- a/tests/test_litellm/proxy/proxy_server/test_routes_audio.py
+++ b/tests/test_litellm/proxy/proxy_server/test_routes_audio.py
@@ -86,6 +86,24 @@ def patched_speech_error(monkeypatch):
     yield
 
 
+@pytest.fixture
+def patched_speech_provider_rejection(monkeypatch, patched_speech_error):
+    import litellm
+
+    async def _raise(*args, **kwargs):
+        raise litellm.BadRequestError(
+            message=(
+                "Gemini TTS only produces raw PCM16 audio, so response_format='mp3' is not supported."
+                " Supported response formats: pcm, wav."
+            ),
+            model="gemini-3.1-flash-tts-preview",
+            llm_provider="gemini",
+        )
+
+    monkeypatch.setattr(proxy_server, "route_request", _raise)
+    yield
+
+
 @pytest.fixture
 def patched_transcription(monkeypatch):
     router = MagicMock()
@@ -198,6 +216,18 @@ def test_audio_speech_error(client, auth_as, patched_speech_error, path):
     assert len(response.content) > 0
 
 
+def test_audio_speech_bad_request_maps_to_400(client, auth_as, patched_speech_provider_rejection):
+    """Regression for LIT-6501: a BadRequestError from the speech path surfaced as a generic 500."""
+    payload = {"model": "gemini-tts", "input": "Hi", "voice": "Kore", "response_format": "mp3"}
+    with auth_as():
+        response = client.post("/v1/audio/speech", json=payload)
+    assert response.status_code == 400
+    error = response.json()["error"]
+    assert "response_format='mp3'" in error["message"]
+    assert "pcm" in error["message"]
+    assert "wav" in error["message"]
+
+
 @pytest.mark.parametrize("path", ["/v1/audio/transcriptions", "/audio/transcriptions"])
 def test_audio_transcription_happy_path(client, auth_as, patched_transcription, path):
     """Pins ``POST /v1/audio/transcriptions`` / ``POST /audio/transcriptions`` (happy)."""