The feature extraction is a very important step in the music audio classification. This task has been performed by renowned descriptors using, in most cases, the time-frequency approach. In this article we propose a descriptor that performs the feature extraction in a set of music audio files labeled in symphonic and percussive music, using parameters calculated within the Euclidean domain. First we calculate the variance fluctuation series of music signal, after we map this series into visibility graphs [13]. At the end each audio track will correspond to a network, where the links are defined by the visibility of variance fluctuations of their respective audio signal. Then, we measure the strength of the partitions of each network in clusters, using calculation of modularity. The results of computation of this parameter in sixty networks showed that percussive and symphonic music can be distinguished and hierarchized on a growing rang, following a direct correlation with modularity.
CITATION STYLE
Melo, D. de F. P., Fadigas, I. de S., & Pereira, H. B. de B. (2017). Community detection in visibility networks: An approach to categorize percussive influence on audio musical signals. Studies in Computational Intelligence, 693, 321–334. https://doi.org/10.1007/978-3-319-50901-3_26
Mendeley helps you to discover research relevant for your work.