speaker-segmentation-fine-tuned-bengali

Speaker Segmentation Fine-tuned for Bengali

This model is a fine-tuned version of pyannote/segmentation-3.0 for Bengali speaker diarization.

Model Description

Fine-tuned on Bengali conversational data for improved speaker segmentation performance.

⚠️ Setup Required

You need to add a config.yaml file to use this model.

Copy the config from pyannote/segmentation-3.0/config.yaml and upload it to this repository.

Usage

After adding config.yaml, load the model:

from pyannote.audio import Model

model = Model.from_pretrained("lucius-40/speaker-segmentation-fine-tuned-bengali")

Or use in a diarization pipeline by updating the pipeline's config.yaml:

pipeline:
  params:
    segmentation: lucius-40/speaker-segmentation-fine-tuned-bengali

Training Details

• Base Model: pyannote/segmentation-3.0
• Language: Bengali (bn)
• Training Epochs: 3
• Batch Size: 2 (effective: 16 with gradient accumulation)
• Learning Rate: 1e-3
• Dataset: Custom Bengali speaker diarization dataset

Files in this Repository

• pytorch_model.bin: Fine-tuned model weights
• config.yaml: ⚠️ NEEDS TO BE ADDED - Copy from pyannote/segmentation-3.0

Citation

@misc{speaker-segmentation-bengali-2026,
  author = {Lucius},
  title = {Speaker Segmentation Fine-tuned for Bengali},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/lucius-40/speaker-segmentation-fine-tuned-bengali}}
}