whisper-tiny-en-US

This model is a fine-tuned version of openai/whisper-tiny on the PolyAI/minds14 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5454
• Wer Ortho: 0.3054
• Wer: 0.3028

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 50
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
3.1254	1.0	14	2.3537	0.5256	0.4044
1.5373	2.0	28	0.8677	0.4152	0.3861
0.4607	3.0	42	0.4872	0.3714	0.3678
0.232	4.0	56	0.4721	0.3325	0.3306
0.1367	5.0	70	0.4858	0.3239	0.3247
0.0665	6.0	84	0.4996	0.3251	0.3235
0.0314	7.0	98	0.5340	0.3208	0.3205
0.0104	8.0	112	0.5454	0.3054	0.3028
0.006	9.0	126	0.5581	0.3152	0.3135
0.0034	10.0	140	0.5593	0.3146	0.3129
0.0024	11.0	154	0.5686	0.3041	0.3028
0.002	12.0	168	0.5736	0.3054	0.3040
0.0019	13.0	182	0.5731	0.3048	0.3034
0.0021	14.0	196	0.5729	0.3054	0.3040
0.002	15.0	210	0.5730	0.3054	0.3040

Framework versions

• Transformers 4.48.0
• Pytorch 2.6.0
• Datasets 3.6.0
• Tokenizers 0.21.4