Fact · Languages and accents
25.2BLEU points
On the CoVoST 2 test of speech translated into English, zero-shot Whisper scored 25.2 BLEU for low-resource languages.
Checked against its source on October 17, 2026
In the source's own words
“Maestro 38.2 31.3 18.4 25.2 Zero-Shot Whisper 36.2 32.6 25.2 29.1”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Table 4, X to English speech translation.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- This is the low-resource column of the same table. Whisper scores higher here than the older systems listed in the table, which the paper attributes to its large pre-training set.
- What it does not show
- The low tier is an average over several languages, so it hides large differences between them. The Yoruba and Swahili results on FLEURS show how far individual languages can fall below the average.