Whisper API returns 524 error on .mp3 files larger than 12 MB

API Bugs
1

Just curious if anyone else is having this issue. Been happening since yesterday afternoon on code that I wrote for my mother’s practice, which has been working for ~4 months without this issue. Key details:

Requests hang for a couple minutes then return the following error:
openai.InternalServerError: Error code: 524 - {‘error’: {‘code’: 524, ‘message’: ‘Gateway timeout.’, ‘param’: None, ‘type’: ‘cf_gateway_timeout’}}

Code:

from openai import OpenAI
import sys
OAI_client = OpenAI()
audio_file = open(sys.argv[1], "rb")
transcript = OAI_client.audio.transcriptions.create(
    model="whisper-1", 
    file=audio_file, 
    response_format="text"
  )

print(transcript)

This is what im running in my git bash terminal

py test.py "audio_files/bigsample.mp3"

Seems like they may have inadvertently lowered the file size limit or something. Please let me know if you’ve experienced anything similar. Thanks.

2

This is happening for me for smaller files as well, works for some, not others. Really want to get a resolution from this or hear others’ suggestion for workaround; my production app depends on it.

3

@elin44 for a workaround: I use SYSTRAN/faster-whisper (on GitHub) to transcribe locally. Same accuracy as the online API, but MUCH slower and memory-hungry. But it could be a (slow) workaround while this 524 issue gets fixed (I also get the error on 18MB file). You could use the official openai/whisper as well, but faster-whisper is more performant. Hope it helps.

4

Yeah this is really bad it’s failing for me and my biggest customers are complaining…

5

Thanks all for the confirmation. I tested it this morning, and the issue seems to be fixed.