The speech recognition API in the MacOS does this in what I think is the most logical way -- you give the software a command dictionary of words or phrases that it should understand, and it just lets you know when it hears one of the commands it knows about. Works pretty well, too.

IMHO, this is the way the Empeg SHOULD work, since this would prevent having to record each of your playlist names for it to compare -- it should know your playlists, and be able to do speaker-independent recognition based on that. Again, since you're not doing continuous dictation of arbitrary works, reliability is good.

Alex