Modern computing has taken over just about every aspect of our music, from creation to production. With programs like iTunes, Garage Band and Virtual DJ, the possibilities for making and organizing our favorite tunes are endless. Engineers at the University of San Diego have found a way to teach computers how to label songs with examples submitted by users, called “game-powered machine learning.” The researchers hope this technique will eventually lead to a text-based multimedia search engine that will find all kinds of music floating around on the Web.

BLOG: Making Music With the Sounds of Symmetry

For the study, researchers had computers learn examples of music that were labeled by fans through a Facebook game called “Herd It” to get category examples. The game had users listen to songs and tag them with categories they felt fit best. The tag with the most votes was the one the computer recognized. The computer analyzed waveforms and recorded common acoustic patterns. This, along with the user-provided tags, made it possible for the machine to predict what songs went with what keywords.

According to the findings, published in the Proceedings of the National Academy of Sciences, music lovers would be able to find any song on the Web by searching with keywords like “funky” or “happy,” even without the aid of track titles or artist names. That could give us a new way to find our favorite songs we never heard of.

Credit: Digital Vision / Getty (top); Jacobs School of Engineering (bottom)