Engineers have developed an artificial intelligence system that lets someone wearing headphones look at a person speaking for three to five seconds to 'enroll' them. The system then plays just the enrolled speaker's voice in real time, even as the pair move around in noisy environments.