Views
No views yet
pyannote/segmentation-3.0 fine-tuned on a combination of competition-provided
audio files (10 hours) and streamed multi-speaker talkshows (5 hours) tailored
to robustly recognize Bengali acoustic spaces, overlap, and phonetics.1from pyannote.audio import Pipeline, Model
2from huggingface_hub import snapshot_download
3
4# Download your model snapshot
5ckpt_path = snapshot_download("rasel-harun/GMPropyannote-bengali-finetuned-10-subset-file")
6
7# Load standard diarization pipeline
8pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization-3.1", token="YOUR_HF_TOKEN")
9
10# Swap segmentation head with your newly trained Bengali-adapted module
11seg_model = Model.from_pretrained(ckpt_path)
12pipeline._segmentation.model = seg_model