GOING INSIDE THE MUSIC: To find the songs you don't know you like. |
It’s a cliché by now, but the Internet allows you to be whomever you want to be. It’s that singular cultural realm where your self-image can relax, unfettered by the perceptions of friends, family, or strangers. Rupert Murdoch might be making millions off your MySpace account, but he has almost no say in what you turn it into. You can sum up your unknowable self in a ten-word headline, or thousands of words of blogging each day, or a picture of your dog. Whatever you choose, you have effectively immortalized your personality. Until you decide to change it next week.
Most Web sites that engage in automated music recommendations try to cater to whatever your Internet image is today. Amazon.com suggests albums based on customers with search patterns similar to yours. Music-review Web sites function as cliques, by establishing artists that represent their audiences, and basing future reviews on the images they’ve created for themselves. Last.fm matches music your computer has played with the listening habits of other users. If your tastes seem similar, they suggest songs you don’t own that like-minded users do.
A culture of “if you like that, then you’ll like this” recommendations occasionally yields new treasures, but at the same time it’s a bit stifling and simplistic. Just because I revere Radiohead doesn’t mean I’m going to love an icy Thom Yorke solo album, and just because I like MF Doof damn well doesn’t mean I’ll dig Gnarls Barkley. These socially constructed suggestions are generally obvious and accurate, but they serve to pigeonhole you into a scene you’re probably already a part of. Your musical taste doesn’t grow like a family tree; it’s just a random accumulation of songs and artists that speak to you in whatever ways you want to be spoken to. It’s a representation of the dynamic you that no one else understands. Isn’t it?
But Pandora doesn’t think your tastes are all that random. Like Last.fm and other sites, the site proudly argues that its developers have come up with a foolproof way to help you find new music you’ll like. Pandora’s approach, though, is completely divorced from what your friends like, what’s selling well, or even what a band’s songs are about. Instead, Pandora’s people examine the sonic makeup of every individual song on their database, and attempt to offer what those other sites can’t: a deductive, logical explanation of what you like.
Pandora is an Internet music-playing site created by the technicians at the Music Genome Project. The project started in California in January 2000, as a group of musicians and like-minded tech-geeks set a lofty goal for themselves: to create “the most comprehensive analysis of music ever.”
Comprehensive comprehension
It works like this: the dozens of musical scientists working with the Music Genome Project came up with a laundry list of hundreds of sonic and lyrical qualities that could technically define, as it were, the genetic makeup of a song. (You can view the list — from “acid jazz roots” through “twelve eight time signature” to “wet snare” — on Wikipedia’s “Music Genome Project” entry.) One of Pandora’s formidably trained technicians — usually with a degree in music studies, trained for 150 hours before being put to work, and paid handsomely for part-time hours because of the tedious nature of the job — spends about twenty minutes with every song selected for inclusion in Pandora’s library. They assign it all of the appropriate qualities from the MGP’s master list, upload it onto the Pandora database, and it becomes yours for the discovering. Pandora selects songs for the database subjectively: they listen to everything their sent, but are bound by the limits of time and 40 years of popular music to catch up with while adding recent artists at the same time.
