Michael Mandel - "MajorMiner: Automatically describing music"

Michael Mandel is a Doctoral student in the Department of Electrical Engineering, Columbia University.

In order to computationally describe music, objective and specific descriptions of the sound must be collected to serve as training data. MajorMiner is a web-based game that makes collecting descriptions of musical excerpts fun, easy, and useful.  When enough instances of a description have been identified (between 20 and 40), we can train a machine learner to automatically identify it.  In evaluations, MajorMiner's top 25 tags were correctly identified 67.2% of the time, while tags from a popular social music website were correctly identified 62.6% of the time.  I will demonstrate the game itself, discuss the collected data, and demonstrate clip retrieval and "semantic music similarity" using these automatically generated tags. You can check out the game and search at http://majorminer.org