Music Type Groupers (mtg): Generic Music Classification Algorithms

This paper outlines our submissions to different music classification tasks for the Music Information Retrieval Evaluation eXchange (MIREX) 2009. We detail here three different algorithms tested in mood and genre classification tasks, and in classical composer identification. These algorithms are based on Support Vector Machines, Disjoint Principal Components Models, and RCA-kNN. The last one utilizes Euclidean distances in a reduced space using Relevant Component Analysis and Kullback Leibler divergence on Mel Frequency Cepstrum Coefficients (MFCC). 1. FEATURE EXTRACTION The submissions are coded in C++ and python. For the feature extraction part, we use an internal library of the Music Technology Group called Essentia 1 . This library contains the features outlined below. We divide our features in two main categories. The “base” features which are state-of-the-art MIR features and the “high-level” features. We aggregate frame-based descriptions using mean and derivatives until second order, variance and derivatives until second order, minimum, and maximum.