openai/whisper-base-Ko

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

• Loss: 0.0970
• Wer: 6.5502
• Cer: 2.9012

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: 5e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3775	0.72	500	0.2690	22.8580	8.2443
0.1316	1.44	1000	0.1760	15.9012	6.8624
0.0658	2.16	1500	0.1285	10.6761	4.2753
0.0273	2.87	2000	0.1133	10.6309	5.0251
0.0112	3.59	2500	0.1040	8.0560	3.3448
0.0055	4.31	3000	0.1010	7.3633	3.2389
0.0024	5.03	3500	0.0979	6.6105	2.9837
0.0013	5.75	4000	0.0967	6.7309	2.9680
0.0009	6.47	4500	0.0967	6.6707	2.9405
0.0008	7.18	5000	0.0970	6.5502	2.9012

Framework versions

• Transformers 4.33.2
• Pytorch 1.12.1
• Datasets 2.14.5
• Tokenizers 0.13.3