How to get the calculated embedding for each segment from the pipeline output ?

[amitli1]

Is there a way to get the calculated embedding for each segment from the pipeline result ?

from pyannote.audio                    import Pipeline
pipeline   = Pipeline.from_pretrained("pyannote/speaker-diarization")
diarization = pipeline("test.wav")
# how to get the embedding for each segment from diarization ?

How can I get the caluclated embedding for each segement ?

I know I can iterate over (diarization.itertracks) and get the relevant segment times and recalculate the embedding (pyannote/embedding), but it seems redundant because this information has already been calculated.

[KarimMibrahim]

Did you find a solution?

[amitli1]

No

[hbredin]

The solution you (@amitli1) propose is actually the only one I can think of.

Indeed, separate embeddings for each segment are in fact never computed during inference.
I suggest you read this paper (and more precisely Section 2.2) for a better understanding of how the pipeline works.