A new technology could let your computer recommend new music you might like based on an acoustic analysis of the tunes it already knows you enjoy.

By analysing the characteristics of a song – like timbre, rhythm, tempo and chord changes – then comparing it to a database of a million songs, the software can recommend similar pieces of music, and even rank them by characteristics, like their key or dance-ability.

"The goal is to be able to have as complete a description of a piece of music as a human being can do," says Xavier Serra, director of the Music Technology Group at the Universitat Pompeu Fabra in Barcelona, Spain. "That's beyond the current state of our knowledge. But what we can do now is enough to develop music recommendation systems based partly or totally on the automatic generation of descriptors."

The university is one of five partners in a European Union-funded project called Semantic Interaction with Music Audio Contents. SIMAC has a presentation of its technology online (1.7MB Flash movie), and is licensing it to private companies.

Losing tracks

One SIMAC partner, Royal Philips Electronics, is developing an MP3 player that uses the technology. The device would help users classify their own music collections.

Serra says that once a collection is beyond about a thousand songs, people lose the ability to keep track of what they have. An automatic classification system would let them describe the type of music they are looking for, and then find it in their own collection.

A number of academic researchers around the world are working on similar systems. Many of them met in September at the International Conference on Music Information Retrieval at Queen Mary, University of London, UK. And at least one company, called Predixis, already has an automatic music analyser on the market.

Web crawling

Serra says SIMAC was able to train on a database of a million songs from Sony. For its music recommender, it combines acoustic analysis with a data mining engine that crawls the web looking for music reviews and other textual information that might also give clues to what the piece of music is like.

Dan Ellis, a computer scientist at Columbia University in New York, US, who does similar work, says that SIMAC is one of the leaders in the field. But despite a lot of academic interest, commercial applications are still relatively unsophisticated.

"If the task is to distinguish classical music from rock, machines can do that pretty well. If you're otherwise going to have to manually sort through them, that's a big win," he says. But more work needs to be done in areas like identifying individual artists, or generating playlists that an individual user will like, he says.