Faster-Whisper output "None" — log 400 "Bed request"
Hi! I’m using the prebuilt Faster Whisper serverless endpoint to transcribe audio files. I send a transcription request using a direct-download Google Drive URL, and the job completes (status COMPLETED) but returns no transcription text—the output field is None. The logs show repeated 400 “Bad Request” errors when trying to return job results. It works locally, but it appears the prebuilt container isn’t packaging the transcription result properly or in a different format. Any ideas what could be causing this? Thanks!
5 Replies
Unknown User•8mo ago
Message Not Public
Sign In & Join Server To View
Debug Report: Faster Whisper Pre-Built Container
(RunPod serverless endpoint: 5i9svwhqcoizei)
Logs & Errors
2025-03-16 12:26:36.406 | error | wqnoc2kcx8qtzf | Failed to return job results. | 400, message='Bad Request'
2025-03-16 12:26:36.354 | info | wqnoc2kcx8qtzf | Started.
2025-03-16 12:26:36.354 | info | wqnoc2kcx8qtzf | Finished.
...
(Repeated “Failed to return job results. | 400, message='Bad Request'” lines)
These errors appear each time the container tries to return the final transcription data.
⸻
Results from Python Debug Script
1) Job Submission:
payload = {
"input": {
"audio_url": "https://github.com/runpod-workers/worker-faster_whisper/raw/main/test/sample.wav",
"model_size": "medium",
"language": "en"
}
}
resp = requests.post(RUN_URL, headers=headers, json=payload)
=> {"id":"b3fbc92d-97f8-4e5e-a115-a68fe83a4be6-e1","status":"IN_QUEUE"}
2) Final Status Check
status_data = requests.get(STATUS_URL + job_id, headers=headers).json()
=> {
"delayTime": 1106,
"executionTime": 62,
"id": "b3fbc92d-97f8-4e5e-a115-a68fe83a4be6-e1",
"status": "COMPLETED",
"workerId": "wqnoc2kcx8qtzf"
}
No "transcription" or "output" field is returned, even though it’s "COMPLETED".
⸻
JSON Parsing Details
The final JSON (parsed in Python with response.json()) only shows metadata:
{
"delayTime": 1106,
"executionTime": 62,
"id": "b3fbc92d-97f8-4e5e-a115-a68fe83a4be6-e1",
"status": "COMPLETED",
"workerId": "wqnoc2kcx8qtzf"
}
No transcription text is present.
Conclusions / Next Steps
1. Jobs always end with "COMPLETED" but the transcript field is missing.
2. The logs show repeated Failed to return job results | 400 Bad Request.
3. I’ve tried:
• Different audio files (including the sample from the GitHub repo).
• Model sizes ("medium", etc.).
• Logging results in Python (json.dumps(status_data)).
• Checking container logs — it says “Started,” “Finished,” then the 400 error.
Questions:
• Is there an environment variable or config needed so the container sends back the text?
• Is this a known bug in the pre-built faster_whisper container?
• Any updated image or extra parameters we can pass to get the full transcription?
Thank you for any guidance!Unknown User•8mo ago
Message Not Public
Sign In & Join Server To View
If you use the prebuilt image then your payload is malformed.Try removing
_url and _size from your keys.Unknown User•7mo ago
Message Not Public
Sign In & Join Server To View