How to pass embedding_exclude_overlap to from_pretrained function ?

[Ca-ressemble-a-du-fake]

Hi,

I want to test [embedding_exclude_overlap](https://github.com/pyannote/pyannote-audio/blob/f700d6ea8dedd42e7c822c3b44b46a952e62a585/pyannote/audio/pipelines/speaker_diarization.py#L75) when doing speaker-diarization to see if it improves the results. I tried to pass it as argument of from_pretrained:

pipeline = Pipeline.from_pretrained("pyannote/[email protected]",
                                    use_auth_token=ACCESS_TOKEN,
                                    embedding_exclude_overlap=True,
                                    )

But it does not work.

I also tried pipeline.embedding_exclude_overlap = True but it silently discards the change.

So how can I use embedding_exclude_overlap ?

Thank you very much for your help!

[hbredin]

pipeline.embedding_exclude_overlap = True should work just fine.
What makes you think it "silently discards the change"?

[Ca-ressemble-a-du-fake]

Thank you for your reply. It made me think the changes were discarded because first I did not see any change in the results after switching to true, and second I tried pipeline.embedding_exclude_overla = True (note the missing 'p' in the end) and it did not raise any error.

[hbredin]

The effect of embedding_exclude_overlap is marginal and it is utterly possible that switching it on or off has no impact on the final output for some files.

Also, embedding_exclude_overlap defaults to True in pyannote/speaker-diarization.

[Ca-ressemble-a-du-fake]

Good to know thank you (I saw False in code, did not think to look at the diarization pipeline).