Whisper large-v2 Korean - ML_project_voice2text

This model is a fine-tuned version of openai/whisper-large-v2 on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0646
• Cer: 1.4722

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: 0.001
• train_batch_size: 4
• eval_batch_size: 2
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 16000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.2332	0.3593	2000	0.2677	7.6162
0.196	0.7186	4000	0.2293	6.0733
0.1524	1.0780	6000	0.1864	5.3510
0.1062	1.4373	8000	0.1448	3.4508
0.0815	1.7966	10000	0.1126	3.2428
0.0349	2.1559	12000	0.0863	1.9394
0.0281	2.5153	14000	0.0732	1.6214
0.0191	2.8746	16000	0.0646	1.4722

Framework versions

• PEFT 0.11.2.dev0
• Transformers 4.41.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1