Research video demonstrating an ability to automatically select individual elements of a recorded song (like the vocal track, guitar solo, ringing cellphone, etc) by singing, whistling, or even Beavis & Butthead style grunting in imitation. Not 100% perfect, but it’s very clever. (I wish the video was embeddable…)
Here’s a link to the original paper (PDF) presented to at the 2009 ACM User Interface Software and Technology symposium.