It's probably in development already, but I think eventually there will be an iPod-like device that lets you select songs with your voice... by singing any part of it, and it could either start from the beginning in one mode or join in where you were singing in another mode. It's not that far-fetched when you think about it. All the component technology already exists, and just needs to be worked together into one unit. Pitch recognition, melody recognition and signature generation. It could even do a combination of voice/pitch recognition so that it can tell what you're getting at by the lyrics alone even if you sing like crap. The lyrics and melody signatures would be easily embedded into music files, provided the artists and record companies don't cry foul... but surely whoever develops this would come up with a proprietary format that protects the information from being abused.
And if nobody's developing it yet, then I claim rights to this idea as of the date of this post and I shall call it iCue. "I Cue" as in it takes your musical cue, and pronounced like "IQ" because it seems intelligent. :)