fine-tuned-Whisper-Tiny-en-US

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

• Loss: 0.7793
• Wer Ortho: 0.3222
• Wer: 0.3247

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 400
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.0014	17.24	500	0.5901	0.3210	0.3188
0.0003	34.48	1000	0.6579	0.3124	0.3142
0.0002	51.72	1500	0.6892	0.3143	0.3165
0.0001	68.97	2000	0.7129	0.3167	0.3194
0.0001	86.21	2500	0.7330	0.3179	0.3206
0.0	103.45	3000	0.7511	0.3191	0.3218
0.0	120.69	3500	0.7653	0.3179	0.3206
0.0	137.93	4000	0.7793	0.3222	0.3247

Framework versions

• Transformers 4.39.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2