akashmjn / tinydiarize

Minimal extension of OpenAI's Whisper adding speaker diarization with special tokens

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Can you please add support for faster-whisper?

hoonlight opened this issue · comments

Thanks for this awesome project.

Could you please add support for faster-whisper?
faster-whisper is a reimplementation of OpenAI's Whisper model using CTranslate2, which is a fast inference engine for Transformer models.

Thank you.

Thanks for the issue.

Would someone from the community be able to help with this? See my comment on #15 for pointers to a reference implementation in python and c++.

Essentially, there is a minor code change needed so that extra tokens aren't suppressed during decoding. The checkpoint structure remains exactly the same. For someone more familiar with that framework, it should be a quick work - hence i'm going to leave this open for a bit.

Feel free to create an issue on faster-whisper and cross reference in case it helps find hands ✋