The Meta Open Source's strongest voice's base model, which supports 1600+ languages

The Meta AI FAIR team released the Omnilingual ASR voice recognition model package, which provides automatic voice recognition in over 1,600 languages with a linguistic error rate of less than 10%; a community-driven design that allows users to extend the model to new languages with a small sample and achieve context-based learning in a large-scale ASR framework for the first time; and an open-source Omnilingual ASR Corpus data set (covering 350 service-deficit languages) and an Omnilingual Wav2vec 2.0 voice representation model with 7 billion parameters。