Machine Learning for Massive Music Search Engine
Researchers at the University of California, San Diego have created a game-powered machine learning system to correctly label the music on the Internet. That is all of the music on the Internet, not just the popular stuff.
The current method for identifying what genre a piece of music involves having paid professionals listen to the music. With new media constantly being uploaded, there are not enough professionals to keep up, and not enough money to pay all of them. This system is meant to replace them though and to be much faster.
The system learns about different kinds of music by using the Herd It game on Facebook, which would be a variant of crowd-sourcing. Machine learning means this system will adapt itself to the information it collects. If it finds a lack of information on jazz, for example, it will change the game’s questions to get the answers it needs. When the system scans music files then, it will compare them against the information it has, to determine the genre or subgenre of music.
Ultimately this could be used to catalog all of the music on the Internet, which would then allow a search engine to analyze the entire collection. This could open up a great deal of the musical world to listeners as all music will be included in the search, instead of just the popular tracks.