Skip to content

Song Science, Part II: Fact and Fiction?


Matt’s post earlier today about Hit Song Science, a piece of software that uses algorithms to analyze songs to predict their likelihood of success in the market, reminded me of a favorite imaginative bit in Jonathan Franzen’s novel The Corrections, which was a big hit a couple years ago:

. . . Brian spent his executive afternoons noodling around with computer code . . . and writing a piece of software that in the fullness of time he quietly patented, quietly found a VC backer for, and one day, on the advice of this backer, quietly sold to the W Corporation for $19,500,000.

Brian’s product, called Eigenmelody, processed any piece of recorded music into an eigenvector that distilled the song’s tonal and melodic essence into discrete, manipulable coordinates. An Eigenmelody user could select a favorite Moby song, and Eigenmelody would spectroanalyze her choice, search a recorded-music database for songs with similar eigenvectors, and produce a list of kindred sounds that the user might otherwise have never found . . . . Eigenmelody was a parlor game, musicological tool, and record-sales-enhancer rolled into one.

Franzen’s fictional concept sounds more feasible and useful than Hit Song Science’s, but the parallels are interesting.

Separate point: I remember reading Franzen’s passage and thinking what an interesting dilemma for a novelist to have: If you come up with a great, compelling idea for an invention, do you pursue it in your art, or in real life?

Posted 11 March 2004