extracting unique speaker embeddings from pipeline

[Aduomas]

have not found a way to extract unique speaker embeddings from the pipeline after running it.

it would allow to similarity function against existing saved speaker embeddings to identify speakers.

[juanmc2005]

Hi @Aduomas, you could get to the centroids (averaged embeddings) of the pipeline in this way:

pipeline = SpeakerDiarization()
# inference over your stream
centroids = pipeline.clustering.centers