Big data and pop music

I’ve talked a bit about the work of computer scientist David Cope who has developed several software tools that compose music. The exact methodology he uses is complex (he’s written several books about it) but his programs have ably output hours of music in the style of various classical masters.

In one of his books, Cope comments that he has not used his software to write pop music. This is partly because he isn’t interested in pop music and partly because he concedes pop music is about a lot more that just the notes on a page (which is what his software is fundamentally creating.) Pop is also about the tone of instruments, their hip factor, and a lot of contextual baggage the performing artists bring to the song (their personal history, persona etc.)

Nonetheless, I think it’s inescapable that computers will be composing pop songs in the future. Or more likely, computers will be helping humans compose pops songs.

But, then what? Cope’s software can generate thousands of variations on a basic tune. Say someone does the same with a pop song. You have 10,000 versions of a certain melody in A minor. Obviously nobody wants to listen to all of them to find the “best one.”

But what if you could look through a data pool of what listeners were listening to and spot upcoming trends? For example, two years ago you could have noted, “Gee, it looks like people are really digging music with these wonky low end gurgles… I bet dub-step will be popular.” Basically, you would note what properties of music seemed to be getting popular and aim the computer composed music towards those styles.

But where would you get this data? This recent NY Times piece, noting that music analysis company Echo Nest has been bought by Spotify, may offer clues.

The Echo Nest is one of a handful of companies specializing in the arcane but valuable science of music data, examining what songs are being listened to by whom, and how. It makes this information available to its clients, including major media companies like Sirius XM, Clear Channel and Univision, which use the data primarily for music-related apps.

“Analyzing music preferences is something we’ve been doing for a long time,” Jim Lucchese, chief executive of the Echo Nest, said in a joint interview with Mr. Ek. “But being directly wired in, and sitting alongside the Spotify team, will give us the ability to push products a lot faster and learn a lot faster than we could before.”

I suspect Echo Nest is, right now, just analyzing “big picture” music trends, like “people are digging hip-hop country songs.” I think eventually they could move towards more granular observations like “major scale melodies that climb high over three bars and then fall down in a giant octave leap in the fourth bar are getting popular,” or “Synth timbres that sound like a theremin and glockenspiel are getting big.” That data could then be used to power the computer aided composition of pop music.

I’m not saying this is a good thing; it worries me. It could certainly lead to an arms race of musical ideas that would result in fads burning out faster and faster. But I think it’s the future.

Leave a Reply

Your email address will not be published. Required fields are marked *