Tim Westergen likes music. A lot. Enough to catalogue just about every song recorded in the past hundred years-and to come up with 400 criteria for categorizing them.
The result, the Music Genome Project, was unveiled in late August with the release of Pandora, which Westergen founded. Located at www.pandora.com, the web program allows users to select a song or artist they enjoy. Pandora then uses the Music Genome Project algorithm to find songs of similar character. It's pretty nifty-though there's no Latin or classical music yet-and the first 10 hours are freewith a $3 monthly subscription charge after that.
recess: So how does the algorithm behind Pandora work?
Tim Westergen: There are close to 400 genes, or musical attributes, that every song gets scored across. To give you an idea, there are 32 "genes" for the voice alone including such genes as vibrato, range, ornamentation, glottal-. We have 30 musical analysts who go through each song-it takes 20 to 30 minutes to do a song. They score them very deliberately across these 400 measurements; that winds up like a musical DNA. When you put in a song for Pandora, it compares that song's DNA to everything else that we have, and it puts together what makes sense musically as a playlist-. If your song's a star, it's finding the other songs in the universe that are close to it. Musically speaking, whenever you interact with [a song], if you give it thumbs up or thumbs down, Pandora understands that as musical feedback-then the algorithm gets better for you.
How did the recording industry react?
There's a couple of answers to that. We launched this service under the guidelines of the Digital Millennium Copyright Act-a legal structure put together by the industry to govern webcasting.
There's a huge amount of commerce that happens because of this. You can link off to buy a song from iTunes or Amazon. It's driving people to a lot of music from things they haven't heard before. Pandora really is a friend to the recording industry because we're helping people find new music.
How did you come up with the idea for the Music Genome Project and Pandora.com?
Ah yes-the moment of insanity. It grew very much out of my own experience. I spent 10 years out of college playing in independent rock bands, traveling around the country, just trying to promote music. I came face-to-face with the challenges of making it as an emerging artist. I got a real deep education.
Then I spent four years as a film composer. My job was to try to find out the desires of a film director. By and large they're not musicians- so I got to thinking about music taste in terms of elements. I had to craft a new composition that they'd like, based on what I knew about their tastes. Those kinds of experiences kind of gelled together over a course of time, and in late 1999 I knew I wanted to find a way to model that knowledge and make it intelligible for an information system. It was when online music was going through the roof.
And how long has it taken to get Pandora running?
Pandora sits on top of the Music Genome Project-the Genome is the guts of the service. Genome's never going to end. That's something we add thousands of songs to every month, which we'll do until people stop writing music, which I hope won't be soon.
So the actual template for the Genome was done in the latter part of 2000, but the Pandora service we launched seven weeks ago.
It seems like you get a lot of repeats on the stations Pandora creates, and it doesn't traverse as many genres as one might hope. Why is that?
The way the service works, if you type in a song, what Pandora is really good at is finding other songs that are similar. The way you get variety is by inputting more than one artist or more than one song on one station-that functionality is not obvious but it's there.
People having been using it in different ways. Some people create lots of stations and flip through them. Other people create a station with more variety within it-. Part of the premise of Pandora is that people like music for different reasons.
How much did it cost to make the Music Genome Project?
Many millions, and we're still spending.
But do you think it will be profitable in the long run?
Well, we don't release the numbers, but it's growing way, way beyond our wildest expectations. We've had an onslaught of users from all over the world-. We could not have hoped for a better start.
Get The Chronicle straight to your inbox
Signup for our weekly newsletter. Cancel at any time.