The gpt-4o-mini-transcribe and gpt-4o-transcribe models transcribe any given input audio file twice, resulting in double outputs. The whisper-1 model does not have this problem.
1 Like
I am experiencing the exact same issue.
No recent deployments of code and it worked fine 12 hours ago.
2 Likes
well…?
It seems that something has changed.
Having the same issue - going crazy combing through logs - I am sure it is not an error on my end… I also can not replicate it though…
Yeah, I believe it was fixed.
Can confirm it has stopped happening on my side as well.
For folks who find this off of search in the future, I found some related issues:
The affected code was open source: KoalaCards/koala/transcribe.ts at main · RickCarlino/KoalaCards · GitHub
Not sure what happened but it certainly seems like an API issue.