Breaking changes in 1.1.0

[jonathanfox5]

I was wondering if there was a list of breaking changes between 1.1.0 and 1.0.3?

I'm trying to get whisperX working with the latest update and it seems that the only change that breaks whisperX is the addition of transcriber.TranscriptionOptions.multilingual. However, I'd be interested if anyone knows of any more that I might have missed?

[MahmoudAshraf97]

There are no breaking changes between 1.1.0 and 1.0.3
WhisperX is not maintained anymore and the latest version known to work out of the box is 1.0.0

[jonathanfox5]

There are no breaking changes between 1.1.0 and 1.0.3 WhisperX is not maintained anymore and the latest version known to work out of the box is 1.0.0

Thanks, I suppose it's a whisperX implementation issue rather than a faster-whisper issue (I previously had to patch whisperx to get it to work with 1.0.3).

My reason for using whisperx is because of the accuracy of the wav2vec2 alignment. Is this something that is on the roadmap for faster-whisper or is there another project that you can recommend?

Thanks for all of your work on this project :)

[MahmoudAshraf97]

For forced alignment, check This which is a lightweight package that offers wav2vec2 alignment for any model and also supports multilingual alignments without needing a model for each language.

[jonathanfox5]

Awesome, thank you so much!