speaker-segmentation-fine-tuned-callhome-jpn

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

• Loss: 1.1719
• Der: 0.2668
• False Alarm: 0.0225
• Missed Detection: 0.0148
• Confusion: 0.2295

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: 0.001
• 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: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Der	False Alarm	Missed Detection	Confusion
0.4449	1.0	44	0.9090	0.2891	0.0225	0.0193	0.2473
0.411	2.0	88	0.9007	0.2767	0.0225	0.0088	0.2454
0.3691	3.0	132	0.8465	0.2570	0.0225	0.0115	0.2229
0.3762	4.0	176	0.8855	0.2585	0.0225	0.0088	0.2272
0.337	5.0	220	0.9608	0.2721	0.0225	0.0142	0.2354
0.3203	6.0	264	1.0052	0.2636	0.0225	0.0152	0.2259
0.314	7.0	308	1.0084	0.2650	0.0225	0.0145	0.2279
0.3066	8.0	352	0.9484	0.2614	0.0225	0.0127	0.2262
0.2968	9.0	396	1.0768	0.2720	0.0225	0.0163	0.2332
0.2847	10.0	440	0.9485	0.2528	0.0225	0.0098	0.2205
0.2784	11.0	484	1.0811	0.2677	0.0225	0.0146	0.2306
0.2674	12.0	528	1.0390	0.2670	0.0225	0.0145	0.2300
0.2646	13.0	572	1.1117	0.2666	0.0225	0.0148	0.2293
0.2425	14.0	616	1.1455	0.2682	0.0225	0.0146	0.2310
0.2569	15.0	660	1.1830	0.2682	0.0225	0.0148	0.2309
0.2497	16.0	704	1.1674	0.2673	0.0225	0.0148	0.2300
0.2494	17.0	748	1.1050	0.2630	0.0225	0.0148	0.2257
0.2334	18.0	792	1.1736	0.2674	0.0225	0.0148	0.2301
0.24	19.0	836	1.1566	0.2679	0.0225	0.0148	0.2306
0.2371	20.0	880	1.1571	0.2650	0.0225	0.0148	0.2277
0.2403	21.0	924	1.1472	0.2640	0.0225	0.0148	0.2267
0.2317	22.0	968	1.1751	0.2676	0.0225	0.0148	0.2303
0.2318	23.0	1012	1.1817	0.2677	0.0225	0.0148	0.2304
0.2322	24.0	1056	1.1723	0.2669	0.0225	0.0148	0.2296
0.2418	25.0	1100	1.1719	0.2668	0.0225	0.0148	0.2295

Framework versions

• Transformers 4.42.4
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1