I wanted some tech like Midomi's ou Soundhound's music fingerprinting mixed in with this. Show me new artists that sound similar to artists that I like. Better yet, similar to a mix of artists that I like. Now that would be nice.
This is actually quite difficult to do. First you need to identify which features of a song are representative of its genre (a song might have 3 million of them). Then you need to build a model that can classify songs accurately based on those features. This has to be done in a speedy way, because you know, you don't have time to wait for a few million songs to process...
Relevant MATLAB code if you want to try your hand: http://labrosa.ee.columbia.edu/~dpwe/resources/matlab/finger...
Relevant algorithm you might want to try (Hidden Markov Model): http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.131...
Company that does recommendations based off audio fingerprinting pretty well: http://www.bmat.com/