bkw1402/KISA-whisper-finetuned

K-Whisper

This model is a fine-tuned version of openai/whisper-base on the 007.저음질 전화망 음성인식 데이터 dataset. It achieves the following results on the evaluation set:

• Loss: 0.5768
• Cer: 24.3026

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: 8
• 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: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
1.9658	0.0328	100	1.5379	37.0310
1.0081	0.0656	200	0.9073	31.1782
0.8246	0.0985	300	0.7621	33.0095
0.7178	0.1313	400	0.6898	44.2204
0.6454	0.1641	500	0.6440	37.7822
0.6279	0.1969	600	0.6134	28.6584
0.5898	0.2297	700	0.5768	24.3026
0.5796	0.2626	800	0.5585	30.3833
0.5415	0.2954	900	0.5514	33.0127
0.5531	0.3282	1000	0.5471	26.6990

Framework versions

• Transformers 4.41.2
• Pytorch 2.7.0+cu128
• Datasets 3.6.0
• Tokenizers 0.19.1