Insanely Fast Whisper
I am trying to get this to work on Runpod to try and eek out some more speed over your faster-whisper which I currently use and love.
It seems like this effort was started? https://github.com/runpod-workers/worker-insanely-fast-whisper/tree/main
@Justin Merrell @Marut
or perhaps its done already and ready to go? I tried rolling my own with the help of an upwork guy https://hub.docker.com/r/joshmohrer/insanely-faster-whisper but its not really working, and somewhat incomplete. I'll be putting growing and hopefully meaningful (to runpod) volume through this, so any help would be greatly apprecaited 🙂
or perhaps its done already and ready to go? I tried rolling my own with the help of an upwork guy https://hub.docker.com/r/joshmohrer/insanely-faster-whisper but its not really working, and somewhat incomplete. I'll be putting growing and hopefully meaningful (to runpod) volume through this, so any help would be greatly apprecaited 🙂
GitHub
GitHub - runpod-workers/worker-insanely-fast-whisper
Contribute to runpod-workers/worker-insanely-fast-whisper development by creating an account on GitHub.
10 Replies
hey @joshmohrer ,
Sure. How can I help? Are you facing any issues with this worker ?
Is that docker image is built using this worker?
Hey @Marut thanks for the reply.
My docker was a clean attempt starting from the insanely-fast-whisper github but its a bit borked.
Is the runpod-workers version linked above something that can be deployed to runpod as-is?
Yeah, You can deploy this worker.
"dt":"2024-02-05 18:21:39.791830"
"endpointid":"3rrxzpnnvgyfyg"
"level":"info"
"message":" "error_message": "Multiple languages detected when trying to predict the most likely target language for transcription. It is currently not supported to transcribe to different languages in a single batch. Please make sure to either force a single language by passing
language='...' or make sure all input audio is of the same language.","
"workerId":"5ohdo2lip7bl7d"
}
your faster-whisper handles this no problem and unfortuntaely its something I need. Perhaps limitation of insanely-faster-whiperthats from your insanely-faster-whiper?
Yeah
I can check once.
i got a bit spoiled by your faster-whisper template which has been really perfect. im partially trying to get diarized output, and partially looking for slightly more speed. my transcription usecase is time sensitive for my app's experience.
I understand, Let me take a look on it.
@joshmohrer made minor changes to be in sync with upstream. You can try with multi language by specifying, now. Let me know if you face any issues.
thanks!!