New model: CrisperWhisper

[imsakg]

I would like to bring attention to the recent release of the new Whisper model, which has demonstrated significant performance on the OpenASR leaderboard. This advancement in ASR technology is noteworthy and could have implications for faster-whisper.

References:
- CrisperWhisper
- OpenASR Leaderboard

[Purfview]

This advancement in ASR technology is noteworthy and could have implications for faster-whisper.

Personally I didn't encountered advancement in any finetuned model.

This one I remember I canceled after the first few segments it outputted, with a thought - "Oh, this not gonna fly."

[Purfview]

Tried it again, on English audio. Bad punctuation, bad segmentation, bad chunking, weird hallucinations with initial prompt, maybe it's not "compatible" with long form transcriptions...