musical signals

Abstract: Searching and organizing growing musical collections for the Algerian radio requires classifying the music signals into a hierarchy of genres to structure them. Musical genres are defined as categorical labels that auditors use to characterize pieces of music. So, a musical genre can be characterized by a set of common perceptive parameters shared by its members. These perceptive parameters are closely related to the instrumentation, rhythmic structure and also harmonic content of the music. An automatic genre classification would actually be very helpful to replace or complete human genre annotation, which is actually used. In this paper, we explore the automatic genre classification of a musical Algerian database. More specifically, two feature sets composed of signal objective descriptors and which are closely related to perceptive ones ( in this case timber) are proposed. The automatic classification of this database is then evaluated through an artificial neural network, more specifically a multiplayer perceptron (MLP). The parameters of this MLP are optimized to obtain the best scores in each case. We thus obtain scores of 60 % to 80 % for eight genres. Interesting comparative results are reported and commented