What is the pipeline used by Speaker-diarization 3.1 ?

[AmirKaseb]

Can i know what techniques or what is the pipeline of Speaker-diarization 3.1 components as i searched a lot and i didnt find a detailed or documentation about the pipeline , even if any one had idea where i can find source code it would be nice

[FrenchKrab]

I think the overall principle is exactly the same as pipeline 2.1 but using different embeddings and a newer segmentation model. So basically you can read the pipeline 2.1 paper, read the powerset segmentation paper, and you should have a good idea of how it works. You can check the pipeline's hparams.yml file on huggingface for parameters details.

I don't know what you mean about the source code since you're on the right repository ... You can start with the speaker_diarization.py pipeline and look for what it makes uses of.