speaker-segmentation-fine-tuned-callhome-jpn

This model is a fine-tuned version of openai/whisper-small on the diarizers-community/callhome dataset. It achieves the following results on the evaluation set:

• Loss: 0.7490
• Der: 0.2217
• False Alarm: 0.0465
• Missed Detection: 0.1331
• Confusion: 0.0421

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Der	False Alarm	Missed Detection	Confusion
0.575	1.0	328	0.7539	0.2338	0.0503	0.1345	0.0489
0.5261	2.0	656	0.7483	0.2256	0.0485	0.1334	0.0436
0.5048	3.0	984	0.7581	0.2248	0.0440	0.1373	0.0435
0.4911	4.0	1312	0.7467	0.2226	0.0472	0.1330	0.0424
0.5161	5.0	1640	0.7490	0.2217	0.0465	0.1331	0.0421

Framework versions

• Transformers 4.40.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1