Fact · Languages and accents
1,107languages
Meta's Massively Multilingual Speech project built a single multilingual speech recognition model for 1,107 languages.
Checked against its source on October 16, 2026
In the source's own words
“a single multilingual automatic speech recognition model for 1,107 languages”
Scaling Speech Technology to 1,000+ Languages, Vineel Pratap et al., Meta AI (arXiv), 2023, Abstract.
Open the original (DOI 10.48550/arXiv.2305.13516)- What it measures
- The model was trained largely on recordings of religious texts read aloud, plus self-supervised learning on unlabeled audio.
- What it does not show
- Research release. Coverage does not mean everyday accuracy: the training text is one genre.