Pyannote Segmentation model fine-tuned on CHiME-7 DASR data

This repo contains the Pyannote Segmentation model fine-tuned on data from CHiME-7 DASR Challenge. Only CHiME-6 (train set) data was used for training while Mixer 6 (dev set) was used for validation in order to avoid overfitting CHiME-6 scenario (Mixer 6 is arguably the most different scenario within the three in CHiME-7 DASR so I used it in validation here as the ultimate score is a macro-average across all scenarios).

It is used to perform diarization in the CHiME-7 DASR diarization baseline. For more information see the CHiME-7 DASR baseline recipe in ESPNEt2 .

Usage

Relies on pyannote.audio 2.1.1: see installation instructions .

from pyannote.audio import Model
model = Model.from_pretrained("popcornell/pyannote-segmentation-chime6-mixer6")