How to feed VAD detection results to diarization pipeline

[kafan1986]

I have already performed VAD step using a different model. If required, I can create a Pynote.Audio.Annotation object from my VAD data. I believe this data more or less represent the segmentation stage output. Is my understanding correct?

Now question is how can I feed this object to embedding and clustering step for performing diarization? I am using the develop branch currently.

[hbredin]

There is no easy way to feed your VAD into the existing diarization pipeline as the latter does not explicitely rely on a VAD step.

You'd have to design your own speaker diarization pipeline. You can use PretrainedSpeakerEmbedding to extract embeddings and then any clustering algorithm from scikit-learn for instance.