Meta's Massively Multilingual Speech Project: Speech-to-Text and Text-to-Speech for 1,100+ Languages!

In a groundbreaking project, researchers have developed speech-to-text and text-to-speech models for over 1,100 languages using wav2vec 2.0 and a new dataset that provides labeled data for over 1,100 languages and unlabeled data for nearly 4,000. The models outperform existing systems and cover 10 times as many languages, making information accessible to millions of people who rely entirely on voice to access information. The dataset was created by collecting audio data from religious texts that have been translated in many different languages and whose translations have been widely studied for text-based language translation research.