Posted on 14 November 2012

Part 2 in a series of videos recorded from ACM MIRUM 2012 in Nara, Japan.

Bob Sturm presents An Analysis of the GTZAN Music Genre Dataset, a detailed investigation into the composition of this popular data set created by George Tzanetakis. By manually inspecting each of the audio tracks in the set, and by using Last.fm and Echo Nest to identify genre information for several of the tracks, he finds that certain flaws in the set might prevent it from training music genre classifiers properly.