Ever have a tune run through your mind, with no name or words attached? When you squawk out what you think might be the melody, people just shrug in perplexity. Now there's a web site that can help.

That's quite an achievement, because music, as Oliver Sacks has pointed out, isn't like a photograph or a PIN code. It's not a thing or a fact, separate from you, to be recollected. Instead, it's an activity you performed once (as a listener or as a player). What we call remembering a song is actually performing the song for ourselves, one more time. Hence, two people's covers of the "the same" tune can be enormously different.

Arefin Huq, Mark Cartwright and Bryan Pardo of the Interactive Audio Lab at Northwestern University have a Web-based solution for you. On their site, Tunebot, you sing your tune as best you can, and their software compares it to a database of other people's contributions. As Christopher Mims describes here, their approach doesn't seek the best match between your yowlings and the songs in, say, the iTunes database. Instead, people who sing to Tunebot are compared to each other: If the site says your tune is probably Take the A Train, it won't be because you came close to some particular recorded version of that song; it will be because you sound like what a lot of other people sound like when they try to sing the song. (In a sense, Tunebot will be creating the musical equivalent of a composite photo: An image that combines many faces to produce an image in which shared traits have been reinforced and individual variations smoothed away.)

Of course, this means that Tunebot will need a lot of examples of people singing, as best they can, the same song. The application, like speech-recognition software, will need to be trained. The Audio Lab people have a clever strategy for that: They will try to get you, the great digital public, to do the work, by turning it into a social game.