Music indexing using independent component analysis with pseudo-generated sources

Abstract

In this paper we present a new approach towards Singing Voice/ Music segmentation using Independent Component Analysis. If the singing voice and the background music are assumed to be two independent signals mixed to form the song, Independent Component Analysis can be used to separate them. ICA requires at least two sources in order to separate two mixed signals, whereas in this case only a single source, i.e. the recording of the song, is available. Another pseudo source is generated from the single source using Discrete Wavelet Transform and the discrimination between singing voice and music is done using a Feed Forward Back Propagation Neural Network. © Springer-Verlag 004.