Automatic musical genre classification using a flexible approach

Autor(es)

Arnal Barbedo, Jayme Garcia ^[1] ; Lopes, Amauri ^[1]

Afiliação do(s) autor(es)

^[1] Univ Estadual Campinas, FEEC, Dept Commun, BR-13083852 Campinas, SP - Brazil

• Tipo de documento: Artigo Científico
• Fonte: JOURNAL OF THE AUDIO ENGINEERING SOCIETY; v. 56, n. 7-8, p. 560-568, JUL-AUG 2008.
• Citações Web of Science: 4

Resumo

A flexible method to classify musical audio signals automatically into musical genres is introduced. The flexibility consists in assigning tip to three genres to a (given signal, taking into account that some musical signals can have elements from more than one genre. The assigned genres are classified according to their influence over the signal. The technique divided the signals into 21.3-ms frames, from which four features are extracted. Some statistical results concerning the extracted features are collected along 1-s analysis segments, resulting in a 12-element vector by segment. Such vectors feed a classification procedure that 1) uses a wide and deep taxonomy. allowing a eticulous Comparison of different genres; 2) performs a pairwise comparison of genres, which allows emphasizing the differences between each pair of genres; 3) assigns, according to the information collected, the degree of influence of each genre over the signal: none, weak, or strong; and 4) performs the final classification, which is given by the genres that best fit the attributes of the signal together with their respective degrees Of influence. This procedure makes it possible to treat adequately the huge amount of songs influenced by or resulting from the fusion of two or more musical genres. The approach has achieved good results with a relatively low computational effort. (AU)

• Processo FAPESP: 04/08281-0 - Classificacao de sinais de audio.
• Beneficiário: Jayme Garcia Arnal Barbedo
• Modalidade de apoio: Bolsas no Brasil - Pós-Doutorado
• Processo FAPESP: 03/09858-6 - Linearidade em sinais, circuítos e sistemas
• Beneficiário: Pedro Luis Dias Peres
• Modalidade de apoio: Auxílio à Pesquisa - Temático