Speaker identification?

[thigger]

Is there any way/plan to support speaker identification in this? The transcription is great but for my use-case I need to know who said what; and it looks like there is a way to support identification in WhisperX. Thanks!

[Pikurrot]

I was planning to do speaker diarization in the long term. Should be possible with whisperx as there is this defined class here. I'm not currently aware whether it's possible with a custom whisper model.
I will try to do that anytime soon, but feel free to contribute.

[thigger]

Thanks. I'll have a look to see if I can help try something. I think I've seen elsewhere people complaining that the diarisation model is gated but I guess having it as an option (with token required) lets people decide.