abhimehra8194/speaker-segmentation-fine-tuned-callhome-hindi

SpeakerSegmentation_Hindi

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: 0.4294
• Model Preparation Time: 0.0006
• Der: 0.1343
• False Alarm: 0.0233
• Missed Detection: 0.0270
• Confusion: 0.0840

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: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Der	False Alarm	Missed Detection	Confusion
0.4597	1.0	194	0.4796	0.0006	0.1635	0.0254	0.0315	0.1066
0.3807	2.0	388	0.4499	0.0006	0.1502	0.0225	0.0312	0.0966
0.379	3.0	582	0.4359	0.0006	0.1400	0.0217	0.0305	0.0878
0.3363	4.0	776	0.4479	0.0006	0.1402	0.0240	0.0278	0.0884
0.3082	5.0	970	0.4358	0.0006	0.1371	0.0245	0.0268	0.0859
0.3125	6.0	1164	0.4287	0.0006	0.1361	0.0214	0.0293	0.0855
0.3143	7.0	1358	0.4247	0.0006	0.1344	0.0233	0.0272	0.0839
0.3081	8.0	1552	0.4211	0.0006	0.1328	0.0230	0.0271	0.0827
0.2999	9.0	1746	0.4298	0.0006	0.1341	0.0233	0.0270	0.0838
0.292	10.0	1940	0.4294	0.0006	0.1343	0.0233	0.0270	0.0840

Framework versions

• Transformers 4.49.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0