whisper-medium-production

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

• Loss: 0.3373
• Wer Ortho: 34.2105
• Wer: 34.2105

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 30
• training_steps: 300

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.0296	20.0	60	0.3520	50.0	47.3684
0.0001	40.0	120	0.3399	34.2105	34.2105
0.0	60.0	180	0.3378	34.2105	34.2105
0.0	80.0	240	0.3370	34.2105	34.2105
0.0	100.0	300	0.3373	34.2105	34.2105

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1