We wanted to optimise for human engagement so we went to real people :) TTS is fine for some things but very few people actually want to listen to a robot reading to them.