speaker-segmentation-fine-tuned-callhome-eng-6

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: 0.5147
• Der: 0.1839
• False Alarm: 0.0668
• Missed Detection: 0.0694
• Confusion: 0.0477

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: 0.002
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20.0

Training results

Training Loss	Epoch	Step	Validation Loss	Der	False Alarm	Missed Detection	Confusion
0.4083	1.0	362	0.4880	0.1967	0.0505	0.0840	0.0621
0.3919	2.0	724	0.4688	0.1852	0.0608	0.0717	0.0527
0.3708	3.0	1086	0.4637	0.1846	0.0581	0.0738	0.0527
0.3549	4.0	1448	0.4636	0.1809	0.0585	0.0689	0.0535
0.3299	5.0	1810	0.4727	0.1835	0.0587	0.0699	0.0549
0.3457	6.0	2172	0.4727	0.1861	0.0654	0.0672	0.0535
0.3241	7.0	2534	0.4921	0.1835	0.0621	0.0701	0.0513
0.3116	8.0	2896	0.4859	0.1839	0.0647	0.0677	0.0515
0.304	9.0	3258	0.4639	0.1788	0.0571	0.0718	0.0499
0.2896	10.0	3620	0.4844	0.1826	0.0659	0.0676	0.0490
0.2853	11.0	3982	0.4696	0.1787	0.0521	0.0777	0.0489
0.2831	12.0	4344	0.4858	0.1831	0.0662	0.0684	0.0484
0.2746	13.0	4706	0.4799	0.1828	0.0639	0.0703	0.0486
0.2685	14.0	5068	0.4951	0.1847	0.0658	0.0695	0.0494
0.2627	15.0	5430	0.5042	0.1829	0.0627	0.0713	0.0489
0.2551	16.0	5792	0.5066	0.1839	0.0671	0.0682	0.0486
0.2509	17.0	6154	0.5126	0.1854	0.0690	0.0695	0.0469
0.2502	18.0	6516	0.5196	0.1861	0.0676	0.0695	0.0490
0.247	19.0	6878	0.5187	0.1844	0.0670	0.0698	0.0476
0.2417	20.0	7240	0.5147	0.1839	0.0668	0.0694	0.0477

Framework versions

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