From 99a6dd02af50e6f822cf57d438a75097232bde15 Mon Sep 17 00:00:00 2001
From: mateo-berri <277851410+mateo-berri@users.noreply.github.com>
Date: Sun, 30 Aug 2026 12:46:11 -0700
Subject: [PATCH] fix(proxy): narrow audio_speech response before reading
 upstream content-type

---
 litellm/proxy/proxy_server.py                 |  5 ++++-
 .../proxy/proxy_server/test_routes_audio.py   | 21 +++++++------------
 2 files changed, 12 insertions(+), 14 deletions(-)

diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 70057201af8..c996735ddbb 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -11068,8 +11068,11 @@ async def audio_speech(
             custom_headers.update(callback_headers)
 
         requested_format: Final = data.get("response_format")
+        upstream_content_type: Final = (
+            response.response.headers.get("content-type") if isinstance(response, HttpxBinaryResponseContent) else None
+        )
         media_type: Final = resolve_speech_media_type(
-            upstream_content_type=response.response.headers.get("content-type"),
+            upstream_content_type=upstream_content_type,
             response_format=requested_format if isinstance(requested_format, str) else None,
         )
 
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 522a20cd34b..de76c7257cf 100644
--- a/tests/test_litellm/proxy/proxy_server/test_routes_audio.py
+++ b/tests/test_litellm/proxy/proxy_server/test_routes_audio.py
@@ -16,6 +16,7 @@ import httpx
 import pytest
 
 from litellm.proxy import proxy_server
+from litellm.types.llms.openai import HttpxBinaryResponseContent
 
 
 @pytest.fixture
@@ -38,20 +39,14 @@ def patched_speech(monkeypatch, request):
 
     monkeypatch.setattr(proxy_server, "add_litellm_data_to_request", _add_data)
 
-    class _FakeBinaryResp:
-        response = httpx.Response(
-            status_code=200,
-            headers={} if upstream_content_type is None else {"content-type": upstream_content_type},
-        )
-
-        async def aiter_bytes(self, chunk_size: int = 8192):
-            async def _gen():
-                yield b"\x00\x01\x02"
-
-            return _gen()
-
     async def _llm_call():
-        return _FakeBinaryResp()
+        return HttpxBinaryResponseContent(
+            httpx.Response(
+                status_code=200,
+                headers={} if upstream_content_type is None else {"content-type": upstream_content_type},
+                content=b"\x00\x01\x02",
+            )
+        )
 
     async def _fake_route_request(*args, **kwargs):
         return _llm_call()