whisper-medium-en-cv-6.3

This model is a fine-tuned version of openai/whisper-medium.en on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 1.0849
• Wer: 30.4960

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: 48
• 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: linear
• lr_scheduler_warmup_steps: 375
• training_steps: 3750
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0	0	2.4579	46.5401
0.7966	0.1	375	1.0410	35.4868
0.5995	0.2	750	0.9551	32.9149
0.3331	1.1	1125	0.9558	32.7312
0.2529	1.2	1500	0.9757	32.3944
0.1245	2.1	1875	0.9818	32.0882
0.1024	2.2	2250	1.0125	31.3227
0.0495	3.1	2625	1.0336	32.0576
0.0438	3.2	3000	1.0665	30.8022
0.021	4.1	3375	1.0777	31.3840
0.0236	4.2	3750	1.0849	30.4960

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.1
• Tokenizers 0.21.1