Centroid Shape

[jakariaemon]

In the earlier version, the centroid shape was 192. However, in Pyannote 3, it's 256.
The pipeline still utilizes the same SpeechBrain embedding model with an embedding shape of 192. Am i missing something here?

[hbredin]

The pyannote/speaker-diarization-3.0 pipeline does not use speechbrain model.

It relies on hbredin/wespeaker-voxceleb-resnet34-LM.

You can still revert back to speechbrain's if needed but you will have to tune the clustering threshold hyper-parameter.

[jakariaemon]

Thank you for the fast response. Solved.