ILSP researchers have developed methods for calculating the similarity between songs based only on the audio content.
The method relies on combining similarities of three main aspects of music, i.e. rhythm, timbre (or alternatively instrumentation) and harmony.
The method is evaluated in MIREX 2013 Audio Similarity Task with encouraging results. More details about the method can be found in:
Gkiokas A., Katsouros V. and Carayannis G., “Deploying Deep Belief Nets for Content Based Audio Music Similarity,” in Proceedings of the 5th International Conference on Information, Intelligence, Systems and Applications (IISA 2014), Chania, Greece, July 2014.
The features extracted from a subset of the Million Song Dataset described in the paper, can be downloaded from the link ftp://media.ilsp.gr/mir/