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: 0.5210
Der: 0.1743
False Alarm: 0.0765
Missed Detection: 0.0670
Confusion: 0.0308

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: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Der	False Alarm	Missed Detection	Confusion
0.6545	1.0	382	0.5338	0.1797	0.0637	0.0805	0.0355
0.6125	2.0	764	0.5097	0.1738	0.0720	0.0692	0.0326
0.6118	3.0	1146	0.5102	0.1709	0.0588	0.0799	0.0322
0.6064	4.0	1528	0.5138	0.1707	0.0657	0.0754	0.0296
0.572	5.0	1910	0.5126	0.1727	0.0709	0.0714	0.0304
0.5671	6.0	2292	0.5161	0.1731	0.0771	0.0654	0.0306
0.533	7.0	2674	0.5143	0.1732	0.0712	0.0715	0.0305
0.551	8.0	3056	0.5207	0.1739	0.0716	0.0717	0.0307
0.5543	9.0	3438	0.5199	0.1738	0.0756	0.0676	0.0306
0.5234	10.0	3820	0.5210	0.1743	0.0765	0.0670	0.0308

Framework versions

Transformers 4.40.0
Pytorch 2.2.1+cu121
Datasets 2.19.0
Tokenizers 0.19.1