Gpt-4o-transcribe returns "audio file might be corrupted or unsupported"

API Bugs
1

Same file is recognized normally when using whisper-1 model. Here’s the file (the signature will expire in 12 hours):

https://bit.ly/4kGstbN

2

The one above was webm file. Now tried with mp3 file - same result: https://bit.ly/41Iwvry

3

+1
i have the same
normal mp3 file - playable on my computer

4

+1 here

is there a guide on the audio file formatting it requires somewhere?

5

+1 here with webm. I tried different options

i.e.:
const options = { mimeType: 'audio/webm;codecs=opus', audioBitsPerSecond: 32000 };

6

Same here :confused:
Works perfectly with whisper-1 in Java and mp3-files, but 400-error with gpt-4o-transcribe …

7

Thanks for reporting! We’re looking into this.

8

This issue should be fixed now. Please let me know if you are still seeing these problems with new the new models on the files that work with the old model.

1 Like
10

Hello, I am encountering the same error with WebM. It appears that the issue has not been resolved.


{
  "error": {
    "message": "Audio file might be corrupted or unsupported",
    "type": "invalid_request_error",
    "param": "file",
    "code": "invalid_value"
  }
}

With :

Content-Disposition: form-data; name="file"; filename="audio-1742599768814-hi2cytp3.webm"
Content-Type: audio/webm;codecs=opus

Tested on files that work with the old model.

12

Do you happen to have request-id returned back in the headers?

13

As a quick workaround: try not passing ;codecs=opus, just pass audio/webm.

14

I can confirm I am having the same issue.

This image shows terminal commands for making API requests to OpenAI's transcription service, with the first attempt using "gpt-4o-transcribe" resulting in an error and the second attempt using "whisper-1" successfully transcribing the audio to "Hello. Hello." (Captioned by AI)
This image shows terminal commands for making API requests to OpenAI's transcription service, with the first attempt using "gpt-4o-transcribe" resulting in an error and the second attempt using "whisper-1" successfully transcribing the audio to "Hello. Hello." (Captioned by AI)1498×563 24.6 KB

Audio files bitrate is 64 Kbps, 1 channel, 48000 sample rate.

15

x-request-id: req_240c6c565993f94e3e425db72894bbfb

{
  "error": {
    "message": "Audio file might be corrupted or unsupported",
    "type": "invalid_request_error",
    "param": "file",
    "code": "invalid_value"
  }
}
16

not working : x-request-id: req_49237981469bbdec426100de7eea7517

Content-Disposition: form-data; name=“file”; filename=“audio-1742602406513-7tp5u4q3.webm”
Content-Type: audio/webm


{
  "error": {
    "message": "Audio file might be corrupted or unsupported",
    "type": "invalid_request_error",
    "param": "file",
    "code": "invalid_value"
  }
}
17

Thanks for sharing the request-id! It seems to be a different issue from the one in the post, which is more specific to a subset of webm files. Would you be open to share the repro input file with me via DMs so I could look more?

18

Yes, I can do that. How can I send you a direct message, please?
I am the developer of Babelfish AI and I am looking to integrate your new models into my extension. It already works very well with whisper-1. I am using this code, which may be useful to you.
webm config is configured here :
github . com jls42/BabelFishAI/blob/main/src/content.js

1 Like
19

Can you please post in this thread when you think the issue is resolved? I’m also getting the same thing and would like to try the fix.

20

I was able to reproduce the issue with webm and I believe it’s now fixed. Please, let me know if you are still seeing these errors.

2 Likes
21

Still the issue is not resolved on different file types.

e.g. request id: ‘req_3033cf2b8e127ad01000af5c2e342d9b’

22

@altyni I think it’s a wav file which is sent with webm extension and it’s expected to receive an error.

1 Like