Fact · Languages and accents
26.1BLEU points
On the CoVoST 2 test of speech translated into English, the Whisper large-v2 model scored 26.1 BLEU for Japanese speech translated into English.
Checked against its source on October 17, 2026
In the source's own words
“Model Arabic Catalan Welsh German Spanish Estonian Persian French Indonesian Italian Japanese Latvian Mongolian ... Whisper large-v2 39.7 31.8 21.5 36.3 40.1 15.0 19.3 36.4 48.1 30.9 26.1 13.9 0.1”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 15 (BLEU scores on CoVoST2), row Whisper large-v2, column Japanese.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- CoVoST 2 is a speech translation benchmark in which speech in each source language is translated into English. BLEU scores how closely a machine translation matches a reference translation, on a 0 to 100 scale, where higher is better. The paper reports the score for each source language.
- What it does not show
- BLEU measures overlap with one reference translation, so it is an imperfect measure of meaning. The score is for read speech in CoVoST 2, not for everyday Japanese recordings, and it is not comparable with word error rates.