Kai He

+1

Did you [accept terms](https://huggingface.co/pyannote/speaker-diarization-3.0#requirements) for the new 3.0 model cards? If 3.0 isn't working for you (it is slower than 2.1 as discussed #499), then try to go back to...

Yes, both segmentation and diarization. You can try my hack in #499 to make 3.0 work fast.

Slightly faster from the limited samples I saw. But I am sticking with 2.1 until no hack is needed for 3.0.

A newer diarization model (3.1) is being used by whisperx. Accept this one instead https://huggingface.co/pyannote/speaker-diarization-3.1