Omnilingual ASR: Advancing automatic speech recognition for 1600 languages
SMRTR summary
Meta's Omnilingual ASR recognizes speech in over 1,600 languages using a 7-billion parameter model, achieving under 10% error rates for 78% of languages. Communities can add new languages using just a few audio-text samples.
SMRTR provides this summary for quick context. The original article belongs to Hacker News.
Read the original article