Musical Bass-Line Pattern Clustering and Its Application to Audio Genre Classification

This paper discusses a new approach for clustering musical bass-line patterns representing particular genres and its application to audio genre classification. Many musical genres are characterized not only by timbral information but also by distinct representative bass-line patterns. So far this kind of temporal features have not so effectively been utilized. In particular, modern music songs mostly have certain fixed bar-long bass-line patterns per genre. For instance, while frequently bass-lines in rock music have constant pitch and a uniform rhythm, in jazz music there are many characteristic movements such as walking bass. We propose a representative bass-line pattern template extraction method based on k-means clustering handling a pitchshift problem. After extracting the fundamental bass-line pattern templates for each genre, distances from each template are calculated and used as a feature vector for supervised learning. Experimental result shows that the automatically calculated bass-line pattern information can be used for genre classification effectively and improve upon current approaches based on timbral features.