We’ve been using Google Music for a couple of weeks now. There’s a lot to like about it. We’ve got more than 50MB of music in it. The UI is responsive and easy to use, especially the Android one. I really like the uploader. It’s just pointed at a directory tree waiting for changes. Whenever I rip a CD or buy an MP3 from Amazon, it’ll soon appear in the cloud with no fuss or clicky-clicky. It’s been working well enough that we’re about to box up all of the CDs and move them to the cellar. None of the others we’d tried were good enough to consider doing that.
We’ve been using the Xoom in a dock as our primary music source for the stereo. One feature which has turned out to be really nice is a button labeled Instant Mix. You just select a track, press Instant Mix, and it creates a playlist with “similar” music. We’ve seen this sort of thing before – for example, we use Pandora on our Chumby. But Google’s really seems to work better than the others. I’ve taken to giving it really odd tracks just to see what it does. It almost always seems to choose something reasonable. The only weakness I’ve found so far is that the uploader slurped up a bunch of Christmas records, and Instant Mix will sometimes try to throw some of those into the mix. That’s just not right this time of year, but I can see why that’d be difficult for it to understand.
Anyways, I’d been wondering how it was implemented. Well, this morning, Google’s research blog had a short article about it. It appears to be a lot more automated than Music Genome (the technology which underlies Pandora). There are a couple of links to interesting papers at the end of that blog post.