I've been using the web since Netscape 1.0, and I've grown pretty jaded about new Internet technology. I got sick of beta testing the latest timewasting Internet fad sometime around 2001. As much as I write and think about technology (real and imagined), I've grown pretty reluctant even to try new technologies myself. And unless an application's reasonably polished and somehow delivers concrete value into my life, I won't give it a second glance.
I say all this as a preface to remarking that I tried the Pandora Internet music service today, and it delivers the goods. Its music recommendation technology is very impressive: type in a band name that you like, and it dynamically constructs a playlist of songs resembling the songs by that band. I was in the mood for some sunny pop music, so I typed in Saint Etienne, and instantly got a playlist of tracks I'd never heard (from artists both familiar and not), and that really sound like music that a Saint Etienne fan would like: Club 8, Stars, Magnetic Fields, Kanda, Brazilian Girls, etc.
I'm really curious about the guts of the so-called "music genome project" technology behind Pandora. Whatever it is, it works. If you're puttering around at home and just want to put on some background music for a particular mood, there's no need to construct playlists manually anymore.
Of course, as a hacker, I find the arbitrary restrictions --- you can only skip forward, and only skip a fixed number of times per hour --- mildly annoying. But if you view Pandora as a replacement for musical radio stations, then it's a huge, quantum leap forward.
My friend MS tried out Pandora some time ago. Being the music geek he is, he mentioned the potential for fiddling around with playlists to optimize your experience. But I didn't want another technology that I spend my time fiddling around with; I want technologies that give me maximum value for minimum time investment. Tragically, therefore, I didn't try it out until today. Rest assured that Pandora will repay the most minimal effort with a large payoff.
The technology behind The Music Genome Project is manual, human categorization by music experts. When you see things like "subtle use of vocal harmony", a human decided to tag the song or artist that way. There are hundreds, probably thousands, of such tags. When you thumbs up/down a song, it uses the tags for that song to help guide your next recommendations.
ReplyDeleteAfter you listen for a while, it will throw you what I call a "wild curve ball", to try to get more types of music into your style choices. Tell it no and you're back on the old track; tell is yes, and the shape of your station changes every so slightly.
It's really killer.
The one shortcoming is that, because the music is hand categorized, they have nowhere near the quantity of stuff compared to a service like Last.fm.
Similar but different check out musicovery, for good mood music, it's less specific, presumably because the classification is less analytic but the interface is a visual joy
ReplyDelete