Fact · Transcription
For the Whisper paper's text scoring, the authors put a space between every letter for Chinese, Japanese, Thai, Lao and Burmese, which means they measured character error rate for those languages instead of word error rate.
Checked against its source on October 17, 2026
In the source's own 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, text standardization for multilingual WER.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- The authors describe a standardization procedure applied before scoring. Korean is not on their list, because Korean text uses spaces between words. The FLEURS figure for Japanese in the paper is therefore a character-level measure, not a count of wrong words.