speaker-segmentation-fine-tuned-callhome-eng-5

This model is a fine-tuned version of pyannote/segmentation-3.0 on the diarizers-community/callhome eng dataset. It achieves the following results on the evaluation set:

• Loss: 0.4674
• Der: 0.1833
• False Alarm: 0.0583
• Missed Detection: 0.0725
• Confusion: 0.0526

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: 0.0003
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 20.0

Training results

Training Loss	Epoch	Step	Validation Loss	Der	False Alarm	Missed Detection	Confusion
0.4679	1.0	181	0.4997	0.2011	0.0620	0.0789	0.0602
0.4255	2.0	362	0.4820	0.1948	0.0604	0.0770	0.0574
0.4084	3.0	543	0.4808	0.1920	0.0598	0.0769	0.0553
0.4017	4.0	724	0.4787	0.1906	0.0584	0.0760	0.0562
0.3911	5.0	905	0.4716	0.1885	0.0572	0.0762	0.0552
0.3845	6.0	1086	0.4676	0.1875	0.0618	0.0718	0.0538
0.3877	7.0	1267	0.4682	0.1877	0.0584	0.0739	0.0555
0.3828	8.0	1448	0.4681	0.1849	0.0579	0.0740	0.0530
0.3768	9.0	1629	0.4645	0.1842	0.0581	0.0733	0.0528
0.3697	10.0	1810	0.4662	0.1838	0.0576	0.0734	0.0529
0.3731	11.0	1991	0.4697	0.1852	0.0607	0.0715	0.0530
0.3691	12.0	2172	0.4642	0.1829	0.0572	0.0734	0.0523
0.3663	13.0	2353	0.4701	0.1854	0.0611	0.0708	0.0535
0.3641	14.0	2534	0.4678	0.1835	0.0591	0.0714	0.0530
0.3631	15.0	2715	0.4655	0.1835	0.0583	0.0724	0.0528
0.3685	16.0	2896	0.4693	0.1838	0.0589	0.0720	0.0529
0.365	17.0	3077	0.4675	0.1836	0.0584	0.0724	0.0528
0.3618	18.0	3258	0.4675	0.1834	0.0582	0.0726	0.0526
0.3651	19.0	3439	0.4675	0.1833	0.0582	0.0725	0.0526
0.3583	20.0	3620	0.4674	0.1833	0.0583	0.0725	0.0526

Framework versions

• Transformers 4.40.1
• Pytorch 2.2.0+cu121
• Datasets 2.17.0
• Tokenizers 0.19.1