metadata
datasets:
- Bingsu/zeroth-korean
language:
- ko
metrics:
- wer
Whisper Medium Kr This model is a fine-tuned version of openai/whisper-small on the Bingsu korean dataset. It achieves the following results:
| Step | Training Loss | Validation Loss | Wer |
|---|---|---|---|
| 1000 | 0.1141 | 0.145484 | 10.811625 |
| 2000 | 0.0369 | 0.106371 | 7.72474 |
| 3000 | 0.0153 | 0.095729 | 6.520102 |