New Delhi, Nov. 12 -- Meta's FAIR team unveiled Omnilingual ASR-an open-weight automatic speech recognition suite that supports over1,600 languages, including 500 low-resource languages and a raft of long-tail Indian dialects. The company also released Omnilingual wav2vec 2.0 (scalable to 7B parameters) and an Omnilingual ASR Corpus of transcribed speech in 350 underserved languages, all aimed at making speech AI more inclusive and easier to bootstrap for new languages.

Meta's release bundles three interlocking assets: the Omnilingual ASR models for transcription, an open-weight Omnilingual wav2vec 2.0 speech representation encoder that scales up to seven billion parameters, and a public corpus containing transcribed speech for 350 under...