Every 20-something social circle seems to have one: the music guru who always has the skinny on the greatest band you’ve never heard. Someone like Will Oremus, who friends say is always turning them on to new tunes. The self-described “music fiend” is an encyclopedia of obscure but talented artists, a good guy to know when confronting the ever-expanding universe of music. “It’s hard to keep up with all the new music I might like to hear and which music I actually do like,” says Oremus, ’04, who scours music-criticism websites like metacritic and pitchforkmedia for album reviews, listens to alternative Internet radio stations like woxy and goes to a lot of live shows. And it’s not enough to hear new stuff, he says; you also have to keep an ear cocked for old music that works for you.
If that sounds exhausting, you can fall back on online recommendations like Amazon.com’s “customers who bought X also bought Y.” But when these are based on music you purchased a while ago, they can be disappointing. “What you want to listen to right now may be different than what you listened to in a very different situation a year ago,” says Andreas Weigend.
He should know. As Amazon’s chief scientist, Weigend, MS ’89, PhD ’91, worked to improve the recommendations by incorporating context-specific information, such as items you’ve recently searched for. Now he’s continuing that mission at MusicStrands, one of two Internet companies with Stanford ties that aim to make online music recommendations more useful and targeted.
MusicStrands, based in Corvallis, Ore., and Barcelona, Spain, relies on the collective wisdom of its users to guide listeners toward new favorites. It’s founded on the notion that people organize their music in a way that makes sense to them. Music- Strands analyzes how its members group songs into playlists: music for the office, the gym, the romantic dinner.
Users upload existing playlists or create them from MusicStrands’ 4.7 million-song catalog, which includes classical music, original cast recordings, spoken word, comedy, even children’s songs. Artists or songs that appear together frequently in lists are considered related, and serve as the basis for MusicStrands’ recommendations. The site also encourages users to tag songs with descriptions such as “rainy-day music” or “driving a convertible,” so that others can follow the sonic breadcrumbs. When they hear a sample they like, they can buy the song or album from one of several referral stores. “If you characterize the use of the music right, the effect it has, then that’s pretty much all there is to it,” says Weigend, who trained as a physicist but became fascinated with finding patterns in human behavior.
Oakland-based Pandora Media, on the other hand, works to find patterns in the music itself. The company employs 30 expert “musician-analysts” to characterize nearly 400 attributes for each song in its Music Genome. Founder Tim Westergren, a jazz-trained pianist, says the concept was “to extract what’s in any trained musician’s head and bottle it,” then make it available to everyone via the web. Qualities like instrumentation, melody, harmony, rhythm and form are the “genes” that give a composition its style and sound, and comparing them can provide a path to finding music you like, says Westergren, ’88.
“It was a crazy idea to do something like this using people, because the universe of music is so large,” he says. But for the musician-analysts, it’s a good gig. They work four to five hours a day, or until their ears give out. Most play in bands, so they appreciate the flexible schedule—and the Friday afternoon jam sessions. So far, the musician-analysts have catalogued songs from more than 10,000 artists—everything from bubble-gum pop to jazz—and they add thousands of new titles each month.
In early summer, both Pandora and MusicStrands premiered initial versions of new services. MusicStrands announced MyStrands, free software that works with Apple’s iTunes and makes recommendations in real time based on the song currently playing. The strength of this approach, Weigend says, is that the recommendations are based on music you actually listen to (as opposed to music you put on your playlist to look cool). And, of course, one click takes you to iTunes if you want to buy a recommended song.
Pandora, which historically has provided recommendations through third parties such as Barnes & Noble or AOL Music, previewed a streaming-audio service that allows consumers to tap directly into the Music Genome. Users begin with a “seed,” typing in the name of a song or artist. In a recent demo, I entered The Arcade Fire, an alt-rock group. Pandora searched the database, looking at every gene, and selected songs by fellow indie rockers Reeve Oliver and AM Radio. Then it streamed the full-length songs through my computer speakers. I was pleasantly surprised by the picks.
Using Pandora is like having your own private radio station. All of the music in the genome is covered by a blanket license, and for $36 a year you can listen to as many songs as you like as often as you like. You can create up to 100 different stations and fine-tune them by providing feedback about which picks you liked and which missed the mark. The service is “kind of like your really knowledgeable indie record store clerk,” Westergren says. Only without the attitude.
Indeed, both companies make efforts to include music from relative unknowns. MusicStrands encourages independent artists to submit their albums. Pandora’s full-time researchers are “out scouring the universe for new stuff,” Westergren says. Plus there’s the stuff people send in—towers of milk crates crammed with CDs. The musician-analysts try to listen to all of it out of respect for the artists, he says, but are selective about what makes it into the genome.
After all, most consumers aren’t craving access to huge catalogs of music. “That’s just not the problem. It’s how do you find something you like?” Westergren says. “The tyranny of choice—music’s got that in spades.” But through efficient serendipity, Pandora and MusicStrands improve the odds.
Read a January 2010 update on this story.
GRETA LORGE, ’97, is an assistant editor at Wired magazine.
