Speech recognition is difficult enough, but if you need to recognise the voice of a particular person as well you will require a LOT of signal processing. Couldn't you make the robot respond to, say, tones of different pitch instead? It would be much simpler.