Fact · Languages and accents
The Whisper paper names five languages whose text it scores by character error rate: Chinese, Japanese, Thai, Lao and Burmese. Korean is not on that list.
Checked against its source on October 17, 2026
In the source's own words
“Additionally, we put a space between every letter for the languages that do not use spaces to separate words, namely Chinese, Japanese, Thai, Lao, and Burmese, effectively measuring the character error rate instead.”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Appendix D, text normalization for scoring.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- The normalizer lowercases text, removes punctuation and then splits words on spaces. For the five listed languages it splits every letter, so the score counts characters. Korean is written with spaces between words, so the paper's word error rate applies to it.
- What it does not show
- The authors call their normalizer imperfect and say it sometimes produces unintended outputs. Korean spacing is inconsistent in practice, so a word error rate for Korean also counts spacing mistakes.