Today's lab team: one drummer, two saxophonists, two guitarists, one bassist, two pianists, one violinist. Today's specimen: Eminem's rap anthem "Lose Yourself." Today's mission: crack all 235 of the song's genes.
Welcome to the Music Genome Project, the most awesomely and exhaustively nerdy cataloguing endeavor in pop-music history. Founded by Stanford alum Tim Westergren, a pianist and former film composer, the Oakland-based project breaks down songs according to their component traits, or genes, just as the Human Genome Project has mapped nature's blueprints. "Genes collectively make a person tall or short, black or white, fast or slow, with freckles or not," Westergren says. "It's kind of like your building blocks -- and we think of this as the same thing for music."
"Lose Yourself" thus has a gene describing whether the bass is played ostinato or as a riff, and another for whether the kick drum sound is tight or booming. There are genes for handclaps, turntable scratches, and organ solos -- times 235. In fact, the four genres, or "genomes," these music analysts have scrutinized to date -- jazz, hip-hop/electronic, rock/pop/country, and world music -- contain a total of about four hundred genes. Some are genre-specific -- hip-hop, for example, has no need of the jazz gene that counts improvised sax licks.
This is far from an academic exercise. As chief strategy officer of Pandora Media, a company he created to oversee the project, Westergren is attempting no less than an egalitarian revolution in music marketing -- one he believes can reach a vast, neglected cadre of fans who love music but have aged out of the crate-digging demographic, or are uninspired by MTV. Pandora's goal is to introduce these people to new music based on what they already like, and while it's too early to draw conclusions, the strong initial response to Pandora's unique product suggests its effort, or something similar, could help transform a recording industry battered by illegal downloading, flagging CD sales, and listener ennui.
Westergren's idea was simple: Compile hundreds of thousands of songs into a gigantic database that can match those you love with structurally similar tracks, thereby creating the ultimate music-recommending device. That was Phase One, which took six years to finesse.
Phase Two, which launched last fall, is where Westergren and his colleagues hope to recoup their startup costs and turn their musicology into a profitable online venture. Pandora.com, named in honor of the curious, is the Music Genome Project's public interface. Type in a song or band that you like, and Pandora combs the database to create a personal Web radio station that plays "genetically" similar songs -- ideally, it will churn up tracks you already like and introduce you to some pleasant surprises along the way. Enter "Pixies," and Pandora launches the band's live cover of "Wild Honey Pie," along with a note: "We're playing this track because it features grunge recording qualities, a subtle use of vocal harmony, mild rhythmic syncopation, repetitive melodic phrasing, and extensive vamping."
Maybe "extensive vamping" is a big deal to you, maybe not. The point is that all the songs that follow on your new "Pixies Radio" station should sound pretty familiar without being exactly alike: Mission of Burma, Black Flag, Mudhoney, the Fall, plus a half-dozen bands you're guaranteed to never have heard of. One mouse click will net you a list of the traits that made them musicological dead ringers. Not that you'll necessarily need it: At least some of the matches are so right-on it's uncanny.
Because Pandora's stations are generated on the spot, they represent a radical departure from other Internet radio stations, which often rely on prechosen playlists or are mere simulcasts of terrestrial stations. Pandora's selections are based on sheer musicology, not sales rank or purchases made by like-minded customers, which is how sites like Amazon and Netflix recommend products. "That's a popularity contest," Westergren says. "Our system doesn't actually know how popular a piece of music is. It doesn't know who else liked it. All it's concerned about is the musical content."
In other words, Pandora gives equal weight to megahits, deep-catalogue cuts, obscurities, and indie tracks. "You can be a completely unknown garage band, and if your song is the right recommendation for the latest Mariah Carey hit or Wilco record, then you're going to end up there," Westergren says.
The site has obvious appeal to record geeks, but so far its greatest response -- this is key to Pandora's business model -- has been from people well past their music-buying heyday who are hungry for a pain-free introduction to music they might like. It doesn't hurt that Pandora is free to anyone willing to tolerate a few on-screen ads, or about $3 a month for an ad-free subscription. Since the free version debuted in November (the subscription site launched last August), Pandora has become a tech-blog darling -- it's been Slashdotted (twice), Farked, and made the subject of about three thousand blog entries. The site's execs are loath to reveal subscriber statistics, but insist that site usage is on a sharp ascent: Four million stations have been created to date, three million of them in December alone. As the site's popularity rises, so do its revenues. Besides subscriptions, Pandora earns its money by hosting ads, and via paid links to Amazon or iTunes, where listeners can purchase a song or album directly.
The underlying music genomes are also undergoing a building boom. Pandora employs more than thirty analysts, all trained musicians and musicologists, to fill its database. It typically takes at least fifteen minutes to score a track, ranking each gene on a scale of zero to five. The analysts usually work in libraryesque silence, headphones firmly clamped over ears, plugged into computer banks that randomly serve up a constant stream of tracks to decode. To ensure consistent scoring, they must go through five group-training sessions in order to master a new genome. These are far noisier endeavors filled with good-natured bantering and people playing imaginary tabletop pianos or pounding out rhythms as they deconstruct the music together.
Seven Days - February 24, 5:52 PM
Seven Days - February 21, 3:30 PM
Seven Days - February 18, 5:00 PM
Seven Days - February 17, 4:09 PM
Seven Days - February 16, 3:24 PM