Too bad it couldn't just be a rather simple set of instructions. I'm sure recognizing an artist's name would be difficult, but a set of rather simple pre-known instructions would be fairly simple to deduce with background noise (as I would think. Instead of trying to guess what you were saying, it would obviously just have to determine a 'best match'). It would be pretty neat just to do something like 'Menu, Left, Left, Select, Play, Shuffle Random"