Speech translation systems can be more robust to background noise in real-world settings if they look as well as listen, like humans do. In a new paper, Meta AI researchers introduce MuAViC, the first open multilingual audio-visual corpus for robust speech recognition and translation. This dataset makes it possible to use audio-visual learning for highly accurate speech translation. We used MuAViC to train our AV-HuBERT model, using sound sight, to translate speech in noisy, challenging settings