fix transcription exception handling - /audio/transcriptions

[otaviofbrito]

Title

fix transcription exception handling - /audio/transcriptions

Relevant issues

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have Added testing in the tests/litellm/ directory, Adding at least 1 test is a hard requirement - see details
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🐛 Bug Fix
✅ Test

Changes

Removed the try-except block that was masking the real exception and causing all errors to return a 500 status code.

Before:

After:

200:

[vercel]

@otaviofbrito is attempting to deploy a commit to the CLERKIEAI Team on Vercel.

A member of the Team first needs to authorize it.

[AlexsanderHamir]

Thank you @otaviofbrito, I believe something like this would solve this specific issue:

        except Exception as e:
            status_code = _extract_exception_status_code(e, default=500)
            raise HTTPException(status_code=status_code, detail=str(e))
        finally:
            file_object.close()

def _extract_exception_status_code(e: Exception, default: int = 500) -> int:
 
    status_code = getattr(e, "status_code", None)
    if status_code is None:
        status_code = getattr(e, "code", None)
    
    if status_code is not None:
        if isinstance(status_code, str):
            try:
                status_code = int(status_code)
            except (ValueError, TypeError):
                status_code = default
    else:
        status_code = default
    
    return status_code

[otaviofbrito]

Thanks for the reply, @AlexsanderHamir ! I initially tried a similar approach, but it turns out the logging messages are being duplicated.

Is there a reason for raising HTTP exceptions? When that happens, async_log_failure_event ends up being called twice. Here's what I found:

1. async_log_failure_event is first triggered inside the LLM API call through the hook — this logs the first message.
2. After that, the code hits this section
3. Then it goes here, where self._is_proxy_only_llm_api_error evaluates to true because the error is an HTTP exception, causing it to log again in self._handle_logging_proxy_only_error

The workaround I've found was to avoid raising the HTTP exception. With that, the message is only logged once.

[ghost]

@otaviofbrito why not just re-raise the error, seems like we should maintain the finally block, so we close the file object

[ishaan-jaff]

lgtm