Voiceprint

Every voice has its own mix of pitch, tone and timing. Software can measure that mix and save it as a pattern, roughly the way a fingerprint reader saves the shape of your finger.

Next time you speak, the software compares the new sound against saved patterns and picks the closest match. It is not perfect: a cold, a shout or a noisy room can throw it off.

In XPENG’s demo, the robot is said to combine a voiceprint with facial features and the content of past conversations to recognise someone after they changed clothes.