Can pyannote-audio be set to distinguish the number of people？

[Erwen222]

Tested versions

3.1

System information

Ubuntu 16

Issue description

Sometimes there will be over-segmentation, two people's audio is divided into five people, the longer the audio, the more likely this phenomenon will occur.

Minimal reproduction example (MRE)

wu

[FrenchKrab]

The pyannote-audio pipeline automatically estimates the number of speakers
You can tune the clustering threshold of the pipeline so that new clusters (= speaker identities) are not created as easily.
If this only appear on long audio files the clustering must be at fault.