Latin Musical Genres Intelligent Classication

João Marcelo Abreu Bernardi; Glaucia Maria Bressan

doi:10.20906/CPS/SICITE2015-0045

Resumo

This paper presents a fuzzy system construction for the automatic Latin musical genres classication. Dividing music in genres is an attempt to classify each musical composition by considering objective criteria. The problem with automatic classication processes is that no clear denition and borders can be easily established. Although many techniques have already been proposed, no general solution to the problem exists, mainly due to the imprecise denition of musical genres. The Latin genres to be classied are extracted by Silla Jr et al. (2008) and they are called: tango, salsa, forro, axe, bachata, bolero, merengue, gaucha, sertanejo and pagode. The system inputs features can be split into three groups: beat related (which includes the relative amplitudes and the beats per minute), timbral texture (the first five MFCCs) and pitch related (which includes the maximum periods and amplitudes of the pitch histograms). For each one of the ten musical genres, a fuzzy classication system is constructed for each of the three input groups. The final step of this process, a membership percentage of an instance for each genre is obtained. The increasing number of musical genres, as well as their fusion and the influence that they receive and exert, motivates the use of fuzzy logic, since it is possible to consider uncertainties and the fuzziness among genres boundaries. The results of the classication are promising, as they suggest minor errors and classify data in a way close to the description of human decision criteria.

Palavras-chave: Fuzzy classification; musical genres; neuro-fuzzy systems