Turbo-V3

[brainer3220]

New Whisper model Turbo3 is rolled out.

SYSTRAN/faster-whisper#1025

[philpav]

+1

[lemuriandezapada]

Or even normal large-v3 would be nice to integrate. That one doesn't work either

[brainer3220]

Right. Original model

@lemuriandezapada

[jb892]

Hi @m-bain, any plan to support whisper-large-v3-turbo?

[lemuriandezapada]

@brainer3220 I don't know. I tried importing large-v3 instead of large-v2 and the transcription failed miserably

[wygoralves]

+1

[DaddyCodesAlot]

This project is largely dead, so don't expect many updates, the founder did it as a uni project and has since moved on. With that being said, you can use Whisper-Turbo by justing pulling the model from Huggingface like so:

whisper_model = whisperx.load_model("deepdml/faster-whisper-large-v3-turbo-ct2", device="cuda",download_root='models', vad_options={"vad_onset": 0, "vad_offset": 0},asr_options = asr_options)

I recommend migrating your code to Whisper-Faster, it's implemented most of the features from Whisper-x. If you need Diarization, migrate to RevAI's reverb model.