Monday, August 14, 2006

Gratuitous Plug For Some Friends

Being older than a certain age, I come a little late to the online music party. That is, I have listened to various streaming services over the years (especially Radio Margaritaville), but most just haven't interested me that much. I used to listen to a favorite radio station from my Santa Cruz days, KPIG, but they went to some weird, proprietary Windows software, and I gave up on them.

Then a couple of years ago, a friend pointed me to a beta test of some music technology his company was working on, and I thought it was kind of cool. Now it's been a real service for a year or so, and it's still very interesting.

They're called Pandora, and their shtick is that they have analyzed the heck out of a lot of music, and if you give them a starting point, they'll stream a bunch of music that's related to what you liked. And the results are quite good: they've opted for high-quality audio, which is refreshing.

They've also added an option called "Backstage," where you can get more info on artists, songs, and albums, and search for similar stuff. That's more like what they were doing when I first looked at their technology. It's all based on something they call the Music Genome Project, which is an attempt to break down music into some of it's component characteristics, and then do some fancy matching to figure out what's similar.

Based on my use, they've done a nice job. You can define a number of different "stations" in your profile to suit different moods. It's pretty cool. Go. Try it. It's free (at least for now). They have paid options, too, but if you're willing to put up with a little advertising, you can just listen.

2 comments:

Laura E. Goodin said...

It's quite intriguing, and seems to be somewhat more accurate than Amazon's model.

However, I'm curious about the algorithm. It seems that the more items you rate thumbs-up or thumbs-down, the more you narrow your range of what it thinks you like. Presumably, over time, you'll just get an endless stream of identical songs.

I suppose, though, that it might not add criteria to the search, it would subtract them: for example, if I've got Peter Gabriel on my thumbs-up list, it will look for "male lead singer, subtle harmonies, etc. etc." (which it did for the first few searches). But if I've got Peter Gabriel and the Roches on my thumbs-up list, it will look for songs with the one or two things they have in common (whatever they are, probably the harmonies thing), rather than try to match up all the parameters at once.

Very interesting....

Chard said...

Also, I have a number of rather diverse (I think) "stations", and I don't know whether it takes into account the overlap of them for some kind of overall preference.

Ultimately, they didn't want to use the Amazon model ("people who like this also like this"), because then you could miss out out on non-popular but similar stuff.

I like the notion that it might pluck some ancient, little-known gem out of the pile, just because it has a good harmonica riff or something.

Worth the price of admission, anyway!