fix(proxy): surface client-side (4xx) errors from /v1/audio/speech instead of 500

litellm/proxy/proxy_server.py

@@ -9352,7 +9352,22 @@ async def audio_speech(
         )
         verbose_proxy_logger.error("litellm.proxy.proxy_server.audio_speech(): Exception occured - {}".format(str(e)))
         verbose_proxy_logger.debug(traceback.format_exc())
-        raise e
+        if isinstance(e, HTTPException):
+            raise ProxyException(
+                message=getattr(e, "message", str(e.detail)),
+                type=getattr(e, "type", "None"),
+                param=getattr(e, "param", "None"),
+                code=getattr(e, "status_code", status.HTTP_400_BAD_REQUEST),
+            )
+        else:
+            error_msg = f"{str(e)}"
+            raise ProxyException(
+                message=getattr(e, "message", error_msg),
+                type=getattr(e, "type", "None"),
+                param=getattr(e, "param", "None"),
+                openai_code=getattr(e, "code", None),
+                code=getattr(e, "status_code", 500),
+            )
 
 
 @router.post(

tests/test_litellm/proxy/proxy_server/test_routes_audio.py

@@ -79,6 +79,38 @@ def patched_speech_error(monkeypatch):
     yield
 
 
+@pytest.fixture
+def patched_speech_client_error(monkeypatch):
+    monkeypatch.setattr(proxy_server, "llm_router", MagicMock())
+    monkeypatch.setattr(
+        proxy_server,
+        "proxy_logging_obj",
+        MagicMock(
+            pre_call_hook=AsyncMock(side_effect=lambda **kw: kw["data"]),
+            post_call_failure_hook=AsyncMock(),
+            post_call_response_headers_hook=AsyncMock(return_value={}),
+            update_request_status=AsyncMock(),
+        ),
+    )
+
+    async def _add_data(data, **kwargs):
+        return data
+
+    monkeypatch.setattr(proxy_server, "add_litellm_data_to_request", _add_data)
+
+    async def _raise(*args, **kwargs):
+        import litellm
+
+        raise litellm.BadRequestError(
+            message="OperationNotSupported: speech is not supported",
+            model="gpt-audio",
+            llm_provider="azure",
+        )
+
+    monkeypatch.setattr(proxy_server, "route_request", _raise)
+    yield
+
+
 @pytest.fixture
 def patched_transcription(monkeypatch):
     router = MagicMock()
@@ -162,6 +194,19 @@ def test_audio_speech_error(client, auth_as, patched_speech_error, path):
     assert len(response.content) > 0
 
 
+@pytest.mark.parametrize("path", ["/v1/audio/speech", "/audio/speech"])
+def test_audio_speech_client_error_preserves_status(
+    client, auth_as, patched_speech_client_error, path
+):
+    """A client-side (4xx) provider error on speech must surface as 4xx, not 500."""
+    payload = {"model": "gpt-audio", "input": "Hi", "voice": "alloy"}
+    with auth_as():
+        response = client.post(path, json=payload)
+    assert response.status_code == 400
+    body = response.json()
+    assert "OperationNotSupported" in body["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)."""