A COMPARISON OF CLASSIFIERS FOR MUSICAL GENRES CLASSIFICATION AND MUSIC EMOTION RECOGNITION

摘要

The automatic classification of music into genres is one of the most popular tasks in the music information retrieval research area and it can be framed as a pattern recognition problem. The goal of this paper is to investigate and to apply different classification methods in order to automatically classify Latin musical genres and predominant emotions associated to them. In order to do this automatic classification task, musical physical attributes (sound beats, timbre and frequency) are considered as input of the classification methods and the performances of classification algorithms are analyzed. The classifiers considered in this paper are based on data mining and statistical techniques: Decision Trees, Random Forest, k-Nearest Neighbor (kNN), Support Vector Machine (SVM) and Artificial Neural Network (ANN). Each Latin genre is associated with its predominant emotion, obtained from the literature. Thus, the proposed methodology can be easily used by music therapists in health treatments.