Fact · Languages and accents
5.3percent word error rate
On the Fleurs benchmark, the Whisper large-v2 model had a word error rate of 5.3 percent for Japanese.
Checked against its source on October 16, 2026
In the source's own words
“Model Croatian Hungarian Armenian Indonesian Icelandic Italian Japanese Javanese Georgian Kazakh Khmer Kannada Korean Luxembourgish ... Whisper large-v2 13.4 17.0 44.6 7.1 38.2 4.0 5.3 nan 105.0 37.7 99.7 37.0 14.3 88.0”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 13 (Fleurs).
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- Fleurs is a read-speech benchmark in about 100 languages. The figure is the error rate of the largest Whisper model in the table, run without fine-tuning, in the paper's own evaluation.
- What it does not show
- Read speech in good audio. Real dictation, accents and noise raise error rates. For languages written without spaces the paper reports character-level scoring, and the error rate can exceed 100 percent when the model writes extra text.