refactor(audio-transcription): extract proxy stream helpers; drop unused imports

- proxy_server: extract `_coerce_stream_form_field` and `_wrap_audio_streaming_response` so the `audio_transcriptions` handler stays under PLR0915 (50-statement) limit.
- base transformation: remove unused `AsyncIterator` / `Iterator` imports flagged by F401.
- test: switch `TestProxyStreamCoercion` to import the new helper directly instead of mirroring the inline logic.

Signed-off-by: WaelRabah11 <wael.rabah@multiversecomputing.com>

litellm/llms/base_llm/audio_transcription/transformation.py

@@ -3,8 +3,6 @@ from dataclasses import dataclass
 from typing import (
     TYPE_CHECKING,
     Any,
-    AsyncIterator,
-    Iterator,
     List,
     Optional,
     Union,

litellm/proxy/proxy_server.py

@@ -8028,6 +8028,25 @@ async def audio_speech(
         raise e
 
 
+def _coerce_stream_form_field(data: Dict) -> None:
+    """Form fields arrive as strings; coerce `stream` to a real bool in-place."""
+    raw = data.get("stream")
+    if isinstance(raw, str):
+        data["stream"] = raw.strip().lower() == "true"
+
+
+def _wrap_audio_streaming_response(response: Any, fastapi_response: Response) -> Any:
+    """Wrap a TranscriptionStreamingResponse iterator in FastAPI StreamingResponse
+    so SSE chunks flush directly instead of being JSON-serialized."""
+    if isinstance(response, TranscriptionStreamingResponse):
+        return StreamingResponse(
+            content=response,
+            media_type="text/event-stream",
+            headers=dict(fastapi_response.headers),
+        )
+    return response
+
+
 @router.post(
     "/v1/audio/transcriptions",
     dependencies=[Depends(user_api_key_auth)],
@@ -8056,10 +8075,7 @@ async def audio_transcriptions(
         form_data = await get_form_data(request)
         data = {key: value for key, value in form_data.items() if key != "file"}
 
-        # Form fields arrive as strings; coerce `stream` to bool so downstream
-        # provider transformation sees a real boolean.
-        if "stream" in data and isinstance(data["stream"], str):
-            data["stream"] = data["stream"].strip().lower() == "true"
+        _coerce_stream_form_field(data)
 
         # Include original request and headers in the data
         data = await add_litellm_data_to_request(
@@ -8167,16 +8183,7 @@ async def audio_transcriptions(
         if callback_headers:
             fastapi_response.headers.update(callback_headers)
 
-        # Streaming transcription: wrap raw-byte iterator in StreamingResponse
-        # so FastAPI flushes SSE chunks instead of trying to JSON-serialize.
-        if isinstance(response, TranscriptionStreamingResponse):
-            return StreamingResponse(
-                content=response,
-                media_type="text/event-stream",
-                headers=dict(fastapi_response.headers),
-            )
-
-        return response
+        return _wrap_audio_streaming_response(response, fastapi_response)
     except Exception as e:
         await proxy_logging_obj.post_call_failure_hook(
             user_api_key_dict=user_api_key_dict, original_exception=e, request_data=data

tests/test_litellm/llms/openai/transcriptions/test_audio_transcription_streaming.py

@@ -221,8 +221,8 @@ class TestProxyStreamCoercion:
         ],
     )
     def test_stream_coercion_logic(self, raw, expected):
-        # Mirror the inline coercion in proxy_server.audio_transcriptions
+        from litellm.proxy.proxy_server import _coerce_stream_form_field
+
         data = {"stream": raw}
-        if "stream" in data and isinstance(data["stream"], str):
-            data["stream"] = data["stream"].strip().lower() == "true"
+        _coerce_stream_form_field(data)
         assert data["stream"] is expected