SmartWhatt/rangsitra-whisper-md-cv-plus

Rangsitra Whisper

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

• Cer: 8.1452
• Wer: 28.0668
• Loss: 0.3658

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: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: warmup_stable_decay
• lr_scheduler_warmup_steps: 1025
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Wer	Validation Loss
0.1041	1.0	2052	9.7313	33.2109	0.3496
0.0673	2.0	4104	9.9421	32.1875	0.4641
0.0492	3.0	6156	9.6464	32.3177	0.3956
0.0132	4.0	8208	8.7890	29.2008	0.3397
0.0031	5.0	10260	8.1452	28.0668	0.3658

Framework versions

• Transformers 4.52.4
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4