YAML Metadata Warning: empty or missing yaml metadata in repo card
Check out the documentation for more information.
Pyannote Segmentation Model - Bengali/Multilingual
Fine-tuned version of pyannote/segmentation-3.0 for Bengali and multilingual speaker diarization.
Training Data
- DISPLACE24: 67 recordings (Dev + Eval)
- DISPLACE26: 125 recordings (Hindi)
- Synthetic Bengali V4: 300 synthetic recordings (1-30 speakers)
- Total: 492 recordings
Performance
- Best Validation Accuracy: 76.67%
- Training Epochs: 18
Training Details
- Heavy on-the-fly augmentation (noise, volume variation)
- OneCycleLR scheduler with warmup
- Label smoothing (0.1)
- Gradient clipping
Usage
from pyannote.audio import Model
model = Model.from_pretrained("smam/pyannote-segmentation-bengali-multilingual")
Citation
Fine-tuned as part of DLSPRINT26 Bengali Speaker Diarization Challenge.
- Downloads last month
- 17
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support