JSWOOK/pyannote_3_fine_tuning

This model is a fine-tuned version of pyannote/speaker-diarization-3.1 on the diarizers-community/voxconverse dataset. It achieves the following results on the evaluation set:

• Loss: 0.3134
• Model Preparation Time: 0.0048
• Der: 0.0888
• False Alarm: 0.0134
• Missed Detection: 0.0337
• Confusion: 0.0417

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: 5e-05
• 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: cosine
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Der	False Alarm	Missed Detection	Confusion
No log	1.0	24	0.3180	0.0048	0.0915	0.0119	0.0385	0.0410
0.1903	2.0	48	0.3116	0.0048	0.0903	0.0125	0.0369	0.0409
0.1839	3.0	72	0.3089	0.0048	0.0896	0.0128	0.0357	0.0411
0.1825	4.0	96	0.3176	0.0048	0.0896	0.0131	0.0352	0.0413
0.1797	5.0	120	0.3148	0.0048	0.0892	0.0132	0.0346	0.0413
0.1801	6.0	144	0.3141	0.0048	0.0890	0.0133	0.0342	0.0415
0.1735	7.0	168	0.3137	0.0048	0.0887	0.0134	0.0338	0.0416
0.1705	8.0	192	0.3133	0.0048	0.0887	0.0134	0.0337	0.0416
0.1796	9.0	216	0.3133	0.0048	0.0887	0.0134	0.0337	0.0417
0.1644	10.0	240	0.3134	0.0048	0.0888	0.0134	0.0337	0.0417

Framework versions

• Transformers 4.44.2
• Pytorch 2.5.0+cu121
• Datasets 3.1.0
• Tokenizers 0.19.1