chengyili2005/whisper-small-canto

Whisper Small Canto - Chengyi Li

This model is a fine-tuned version of openai/whisper-small on the Common Voice 24.0 - Cantonese dataset. The following results are achieved on the evaluation set using the best model:

• WER: 62.24
• CER: 12.46

Model description

It's my first time fine-tuning an ASR model.

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Done through Google Colab Pro using the L4 GPU

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• 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: 500
• training_steps: 10000 (about 21-22 epochs)
• mixed_precision_training: Native AMP

Training results

Step	Epoch	Training Loss	Validation Loss	CER
1000	2.1552	0.081	0.2857	13.8249
2000	4.3103	0.0173	0.3094	12.7975
3000	6.4655	0.0039	0.3496	12.7571
4000	8.6207	0.0008	0.3721	12.5457
5000	10.7759	0.0006	0.3784	12.5347
6000	12.9310	0.0043	0.3907	13.0640
7000	15.0862	0.0004	0.4053	12.6560
8000	17.2414	0.0008	0.4123	12.4648
9000	19.3966	0.0002	0.4196	12.4648
10000	21.5517	0.0001	0.4238	12.5071

Framework versions

• Transformers 4.52.0
• Pytorch 2.9.0+cu126
• Datasets 4.4.2
• Tokenizers 0.21.4