One recording drives nine analysed mouth shapes on a shell-fur plush bear, then the same cues drive a robot's screen face. Captions highlight each word as it is spoken. Edit guide