search

pyannote / pyannote-audio

Neural building blocks for speaker diarization: speech activity detection, speaker change detection, overlapped speech detection, speaker embedding
http://pyannote.github.io
MIT License
6.4k stars 789 forks source link

Remove/Exclude overlapping segment for speaker diarization #1780

Open hkpmatt opened 3 weeks ago

hkpmatt commented 3 weeks ago

Tested versions

Tested version speaker-diarization-3.1

System information

window 11, pyannote 3.3.2, python 3.8.10

Issue description

I am using the speaker-diarization pipeline and I have met some problem where there should be no overlapping speech in the audio but the segmentation model seems to falsely detected it. The only parameter for the "pyannote/segmentation-3.0" only have 1 hyperparameter that is the min_duration_off which will not settle the problem. Is there any parameter to remove or exclude the overlapping segment?

Minimal reproduction example (MRE)

NA

hbredin commented 3 weeks ago

There is no such option for now indeed, unfortunately.

hkpmatt commented 3 weeks ago

There is no such option for now indeed, unfortunately.

Does this mean the segmentation model has no option to disable the overlapping speech detection?

hbredin commented 3 weeks ago

As of now, no such option.