---
datasets:
- Bingsu/zeroth-korean
language:
- ko
metrics:
- cer
- wer
base_model:
- openai/whisper-large-v3-turbo
pipeline_tag: automatic-speech-recognition
---

Fine-tuning Whisper :arge v3 Turbo on zeroth Korean dataset. 

Dataset split: 
- test -> 50% validation, 50% test
- Train set duration: 206 hours 43 minutes
- Validation set duration: 2 hours 22 minutes
- Test set duration: 2 hours 22 minutes

Results: 
- validation WER: 4.90%
- validation CER: 1.78%
- test WER: 4.89
- test CER: 2.06