Modelpublic

lucius-40/speaker-segmentation-bengali-optimized-cfg-1-aggressive

sourceHugging Facemitupdated 8mo agoView on Hugging Face

0likes15downloads

Model Card

speaker-segmentation-bengali-optimized-cfg-1-aggressive

This model is a fine-tuned version of pyannote/segmentation-3.0 on the bengali-speaker-diarization dataset. It achieves the following results on the evaluation set:

—Loss: 0.6027
—Model Preparation Time: 0.0046
—Der: 0.1598
—False Alarm: 0.0393
—Missed Detection: 0.0332
—Confusion: 0.0873

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 0.0015
—trainbatchsize: 8
—evalbatchsize: 8
—seed: 42
—gradientaccumulationsteps: 2
—totaltrainbatch_size: 16
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.2
—num_epochs: 10
—mixedprecisiontraining: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Der	False Alarm	Missed Detection	Confusion
0.4211	1.0	508	0.4827	0.0046	0.1617	0.0410	0.0173	0.1034
0.4766	2.0	1016	0.4685	0.0046	0.1602	0.0441	0.0133	0.1028
0.4205	3.0	1524	0.4872	0.0046	0.1548	0.0400	0.0185	0.0964
0.3807	4.0	2032	0.5049	0.0046	0.1558	0.0405	0.0250	0.0903
0.3615	5.0	2540	0.4887	0.0046	0.1477	0.0431	0.0177	0.0869
0.3785	6.0	3048	0.5197	0.0046	0.1570	0.0366	0.0346	0.0858
0.3255	7.0	3556	0.5446	0.0046	0.1573	0.0382	0.0307	0.0884
0.3003	8.0	4064	0.6027	0.0046	0.1598	0.0393	0.0332	0.0873

Framework versions

—Transformers 4.48.3
—Pytorch 2.8.0+cu126
—Datasets 3.6.0
—Tokenizers 0.21.4