Pandora's Long Strange Trip

Tweet

But now, just when Westergren thought he'd finally moved Pandora into safe territory, he's facing his biggest battle yet: steep hikes in music royalty rates that were announced earlier this year. Every time Pandora plays a song, it must pay a small fee to the music's publisher, songwriter, performer, and label. Westergren has lobbied Congress and organized a grass-roots campaign of millions of Internet radio listeners. Why the urgency? The rate hike has sent Pandora's operating expenses soaring and threatens to silence Pandora for good.

Westergren plays the piano, the bassoon, the recorder, the drums, and the clarinet. His personal Pandora stations are based on songs by Muddy Waters, Ben Folds, Josh Fix, Oscar Peterson, Art Farmer, Elvis Costello, and James Taylor. He studied music at Stanford--Stan Getz was one of his professors--and graduated in 1988 with a deep understanding of music theory and computer applications for music, and a notion that he could somehow make money as a musician. He played piano in a series of taking-themselves-seriously acoustic rock bands called Late Coffee and Oranges, Barefoot, and Yellowwood Junction. But after years of driving a van all over the West Coast, crashing in friends' basements, he became frustrated with how hard it was to get noticed. Band life fell apart in 1995.

Westergren started composing scores for low-budget independent films, and that's when he began thinking differently about music. He'd ask directors about the sounds they were searching for and listen as otherwise articulate, creative people struggled to find the right words, usually falling back on descriptions like "something like Natalie Merchant, but more scary." Sitting at his piano, trying to evoke a frightening Natalie Merchant, Westergren thought about what terms such as "scarier" and "darker" and "happier" meant in purely musical terms. Would changing the rhythm, the melody, or the alto sax arrangement produce the desired result? If so, then wouldn't it be possible to create a giant database of music based on its underlying characteristics, which would make it easier for listeners to find exactly what they were looking for?

Around the same time, he read an article about the plight of singer-songwriter Aimee Mann. Though Mann's two previous records had sold a respectable 227,000 copies and won critical acclaim, her record label was refusing to release her current effort; it was focusing instead on blockbuster artists with sales in the millions. For Westergren, Mann's story brought back bitter feelings about Yellowwood Junction, which had built a strong following in the western United States but had no way to get its music out to larger audiences. "All the ideas that had been swimming in my head coalesced at that point," he says.

Lucky for him, it was 1999, and everyone in the Bay Area seemed to be starting a company. Westergren founded his, Savage Beast Technologies, with two friends, Jon Kraft and Will Glaser. Kraft, who'd already started and sold a tech business, helped Westergren sketch out a business plan, Glaser worked on the software, and Westergren worked on the music.

His premise was that music could be broken down in an objective way by trained musicians; he called his effort the Music Genome Project, a riff on the drive to map human DNA. Westergren began by thinking about how any musician would describe a piece of music and came up with about 600 qualities. He hired a professional musicologist to refine the approach and Glaser created an algorithm based on the work. The system now includes about 400 different "genes" for each genre (pop, classical, jazz, and hip-hop) that break down the music's form, the instruments used and their tone (is the saxophone gravelly?), the level of virtuosity, and the overall mood and style. (For classical music, is the mood pastorale, giocoso, or agitato? For jazz, is there an R&B or a smooth jazz influence?) It analyzes lyrical content, along with harmony, rhythm, and pretty much any other factor you can think of. Westergren was insistent on having humans analyze each song, arguing that computers alone would not be able to pick up the nuances and mood. Analysts rate each song on a 10-point scale on each of the 400 attributes. Those ratings constitute the song's musical DNA, which is entered into a database. The goal is to point listeners toward music they might enjoy based on an analysis of what they already like.

The question was how to make money with this thing. In 2000, Westergren pitched the company as an e-commerce site that would recommend music based on the genome, and in March he raised $1.5 million in venture capital. Two weeks later, the stock market crashed, rendering an underfunded e-commerce site a very bad idea. The next four years saw a plethora of different business plans for Pandora. "The first day I started, I sat down with Tim and I said tell me about the product," says Dan Lythcott-Haims, who joined Westergren in 2000 and is now Pandora's creative director. "He said, 'No, you tell us about the product.' There was no product, just a loose idea." Westergren set out to license the genome to other sites as a recommendation engine, but after more than a year his efforts resulted only in a meager $20,000 development fee from Barnes & Noble.com. His next idea was to put the technology into in-store kiosks at music retailers. At the end of 2002, Westergren reached a licensing deal with AOL Music, and Best Buy (NYSE:BBY) chose Pandora over 15 competitors to run a trial kiosk program. Both deals stretched payments out over a few years.

 PREV  1 | 2 | 3 | 4 | 5  NEXT 

Read more:

Sign-up for our Technology Newsletter

Email Address: