Facestar Dataset Description Existing audio-visual datasets for human speech are either captured in a clean, controlled environment but contain only a