Massively Multilingual Speech (MMS) project uses wav2vec 2.0 and a new dataset to support 1,107 languages with speech-to-text and text-to-speech models.
MMS also supports language identification for over 4,000 languages and aims to preserve language diversity and make information accessible.