whisper-new-nnat-10h-maxcos-ecapa

This model is a fine-tuned version of openai/whisper-large-v2 on the JASMIN-CGN dataset. It achieves the following results on the evaluation set:

• Loss: 0.5305
• Wer: 26.4319

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: 0.0001
• train_batch_size: 48
• eval_batch_size: 32
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 91
• num_epochs: 3.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6936	0.2984	91	0.6340	45.6095
0.3958	0.5967	182	0.5687	33.3396
0.3687	0.8951	273	0.5474	28.3426
0.3232	1.1934	364	0.5466	29.8319
0.2992	1.4918	455	0.5458	27.5371
0.3277	1.7902	546	0.5312	26.2352
0.3002	2.0885	637	0.5286	25.6357
0.2726	2.3869	728	0.5317	27.2655
0.2821	2.6852	819	0.5328	26.3991
0.2643	2.9836	910	0.5305	26.4319

Framework versions

• PEFT 0.17.1
• Transformers 4.57.6
• Pytorch 2.8.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2