Classificação automática de emoções em músicas latinas utilizando diferentes fontes de informação

Abstract

With the growing amount of music available online, there has been an increasing expansion in research of musical information and the recovery investigations to automated systems. The Music Information Retrieval (MIR) field looks at various aspects related to how to organize, categorize, and access music collections. The development of new methods and the creation of new musical representations can contribute to the accuracy of classifiers for recognition of emotions, since these are among the greatest challenges in the area of Music Emotion Recognition (MER). This work investigates, implements and combines three different sources of information (cifras, audio and lyrics) for automatic emotion classification in songs. The following activities have been used to develop this work: database definition Multimodal Latin Music Mood Database (MLMMD), preprocessing of the different types of data, mining and combination of the different types of data. Through the procedures applied it was possible to carry out an analysis of the different results. First, it was observed that the multimodal early fusion method was better than the others other approaches (no fusion and late fusion). Second, the Support Vector Machine (SVM) showed an overall average better than the other classifiers.