Voice-controlled apparatus is provided which minimises the risk of activating more than one such apparatus at a time where multiple voice-controlled apparatus exist in close proximity. To start voice control of the apparatus, a user needs to be looking at the apparatus when speaking. Preferably, after the user stops looking at the apparatus, continuing voice control can only be effected whilst the user continues speaking without breaks longer than a predetermined duration. Detection of whether the user is looking at the apparatus can be effected in a number of ways including by the use of camera systems, by a head-mounted directional transmitter, and by detecting the location and direction of facing of the user.