SpeakerDiarizationProtocol stats

[PaulLerner]

annotation_duration should not consider parts that are not annotated right ?
https://github.com/pyannote/pyannote-database/blob/develop/pyannote/database/protocol/speaker_diarization.py#L194

[hbredin]

The current behavior assumes that there is no annotation outside of annotated support.

If there are, this is probably a bug in the provided annotated or annotation values, and it should be fixed upstream.

[PaulLerner]

there is no annotation outside of annotated support

Yes but there might be annotation which we don't want to consider as annotation (as in, e.g. our current forced-alignment setup).
However here you take into account the whole annotation duration, regardless of the annotated uem.
You could fix it by adding .crop(annotated) to annotation = item['annotation']

[hbredin]

I understand the (your :-)) use case and would be happy to merge a pull request that:

• crop annotation by annotated
• warn the user that the crop happened if (and only if) the original annotation was not fully included in annotated
• update the documentation to reflect this breaking change