Tuesday, October 03, 2006
Tim Westergren hangs out at Hampshire--Illuminates the Music Genome Project
It's been and incredibly beautiful day out here in Western Mass--if there's one thing we have, it's beautiful days. Pandora founder Tim Westergren noticed this too...held a meet-up out here at a Hampshire College's Red Barn...
I'm only about 20 minutes away...how could I *not* go....and like everybody else who was there, I was incredibly interested to see what Tim would say about Pandora and the Music Genome Project (Tim was at SXSW Interactive this past March, too--but our paths didn't cross. I did sit in on his panel, which is where I got more interested in Pandora.)
It was great to have Tim out here--scanning his blog tonight, it's easy to see why he'd go to places like L.A. and New York. And he'll be out in Providence, RI on 10/4 for BIF-2, so it would make sense to hold a meetup there.
But out here in what's--technically, and in comparison to Boston--The Middle of Nowhere??
There were about 30 of us Pandora enthusiasts at the Red Barn, and by the questions people asked, everyone had a keen interest in music and a desire to know more about the genesis of the Music Genome Project....a mix of musicians, bloggers, student djs, a couple of reporters, and some curious folks. A good group.
Tim explained that Pandora's main focus is to help independent musicians, and he noted how Pandora takes submissions from "unsigned" bands and weaves them into the Genome--if the music passes a determination that it's of "quality." Now, quality might be a subjective judgment, but Pandora staff doesn't simply judge a piece of music after listening to it for three minutes. Rather, the staff of trained analysts who understand the various elements of music (folks *with* music degrees, who also go thru 150 hours of training by Pandora) listen to a piece of music for between 15 and 40 mins, to determine a piece of music's unique elements. It seemed that Tim was trying to explain that by considering these unique elements, one could determine that a piece of music had a certain kind of "quality."
Listening to Pandora, though, really isn't about tracking down the latest and greatest in Top 40, or the perfect tune; nor is it about being assaulted by all emo-and-garage all the time (that is, unless you program a station for that kind of music.) It's more about discovering new music that fits with music you already like...
Mostly everyone was interested to find out how the Genome works. As Tim explained, there are 400 unique music criteria. Voice alone can be described by over 30 different attributes! As mentioned, people with some expertise and training, listen to the music and mark it for the various criteria, which is then fed into the database. (it was explained how Surowiecki's "wisdom of crowds" idea doesn't quite work for the type of service they wanted to produce--music tastes are too broad to use that particular approach effectively.)
Every song has its own unique mathematical formula--it's own "genes."
Now, when a song or artist is picked out, the magic little computers search out other songs that reflect similarities in this mathematical formula....but it's not always going to pick the exact, perfect song. It needs feedback from us. Then, it becomes very necessary that we give the Genome a bit of feedback by rating a song "thumbs up" or "thumbs down." This way, the Genome can find more music that fits our particular likes and dislikes, not just what it scans and picks for us.
So, two "stations" of the same artist or song can be started at the same time, and each one end up totally different because of the "thumbs up" and "thumbs down" ranking process.
My Led Zeppelin and Ursula 1000 stations are, then, totally different from someone else's Led Zeppelin and Ursula 1000 stations....that is, if there's someone as particular in his/her tastes to also have Led Zep and Ursula 100 as their stations. (as I'm writing this, though, I've been listening to my White Stripes station...)
A funny thing, though, that some of us told Tim about, is that occasionally Pandora hiccups--basically, it can't find the next song in the Genome to play. While they're working on that *not* happeneing, Tim encouraged us to email them when it does. Feedback, after all, is super-important to the whole process.
Two very interesting factoids about Pandora: it grew by word-of-mouth and some props from Slashdot and other tech blogs. And they will never "play a song because someone paid us to." (gee, that beats ClearChannel all to hell...)
I often like to put Pandora to the test. I tend to have some very ecclectic music interests, which sometime befuddle the genome (I think). So, tonight, I entered "Vitas", that weird Russian tenor that looks like a baby Falco and sounds like the spawn of Klaus Nomi. Didn't come up with anything for "Vitas." Wasn't surprised there. So, I entered "Style Council"--Paul Weller's project *after* The Jam. I've come across few folks over the years--other than those who have been into music for as long as I have and were "punks" at one point or another--who know of Style Council. Yet, there they were, in Pandora (Mari Wilson, however, wasn't--but she's a contemporary of Style Council, seriously obscure and British and *very* retro in music style.) Although I'm finding with bands like Style Council, who were a bit tough to classify, that music that corresponds to its "genome" is stuff I really don't enjoy--like "Sweet Spot" from Captain Harry(?!?!eek!). It wasn't a very good fit with Style Council...
I was, though, very pleased with my Erasure station--just perfect for those "Oh! L'Amour!" moments ;-)
Come to think of it, Vitas might fit pretty well with Erasure...