Massively Multilingual Speech (MMS) is an AI speech recognition model open sourced by Meta. Speech-to-text and text-to-speech support for 1107 languages, and speech recognition for over 4000 languages. The MMS project increased the number of languages ​​supported by a factor of 10-40, depending on the task. The principal component is a new dataset based on readings of publicly available religious texts and effectively exploits self-supervised learning. The project team built a pre-trained wav2vec 2.0 model covering 1406 languages, a single multilingual automatic speech recognition model for 1107 languages, the same data…

#Extend #voice #technology #language #MMS

Leave a Comment

Your email address will not be published. Required fields are marked *