Testing supervised classifiers based on non-negative matrix factorization to musical instrument classification

In this paper, a class of algorithms for automatic classification of individual musical instrument sounds is presented. Two feature sets were employed, the first containing perceptual features and MPEG-7 descriptors and the second containing rhythm patterns developed for the SOMeJB project. The features were measured for 300 sound recordings consisting of 6 different musical instrument classes. Subsets of the feature set are selected using branch-and-bound search, obtaining the most suitable features for classification. A class of supervised classifiers is developed based on the non-negative matrix factorization (NMF). The standard NMF method is examined as well as its modifications: the local and the sparse NMF. The experiments compare the two feature sets alongside the various NMF algorithms. The results demonstrate an almost perfect classification for the first set using the standard NMF algorithm (classification error 1.0%), outperforming the state-of-the-art techniques tested for the aforementioned experiment.