whisper-medium.en

This model is a fine-tuned version of openai/whisper-medium.en on an acc_dataset_v3 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4242
• Wer: 0.1293

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

See the training notebook here:

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 4
• 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: linear
• lr_scheduler_warmup_steps: 25
• training_steps: 300
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.6045	2.2727	25	0.9047	0.1535
0.2267	4.5455	50	0.4149	0.1262
0.0207	6.8182	75	0.4454	0.1483
0.0134	9.0909	100	0.4660	0.1388
0.009	11.3636	125	0.4311	0.1462
0.0051	13.6364	150	0.4368	0.1356
0.0018	15.9091	175	0.4294	0.1462
0.0003	18.1818	200	0.4234	0.1356
0.0002	20.4545	225	0.4235	0.1325
0.0002	22.7273	250	0.4239	0.1304
0.0002	25.0	275	0.4240	0.1283
0.0002	27.2727	300	0.4242	0.1293

Framework versions

• Transformers 4.48.2
• Pytorch 2.6.0+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0