The voice recognition can't use AUX - this doesn't go to the CPU. In theory it could be streamed over the serial (pretty darn near 230,400bps as it needs 11025Hz mono 16 bit). It's unlikely we'll make a box to do this, but if someone else did we could most likely support it.

The voice recognition is trainable (for your playlist names) - it isn't syllable based.

Hugo
(Wanting to have a Star Trek set so he can say "Computer!" in a scottish accent and get a response)