Estimating the pitch of polyphonic vocal music using comb filters connected in parallel and singular value decomposition

Abstract

SUMMARYPitch estimation processing is an essential process in the automatic scoring of music. In this paper we propose a method for estimating the pitch of polyphonic vocal lines, an area that has received almost no research. The proposed method estimates pitch by using 12 comb filters connected together in parallel (corresponding to the 12 notes of an octave) together with singular value decomposition (SVD). The approach is based on the fact that a comb filter (H(z) = 1 -z −N ) is able to eliminate all the harmonic components of the note to which it corresponds and also that the SVD of the matrix of signal samples enables us to distinguish the number of harmonic components in the signal from the number of singular values obtained. In other words, together with the above-mentioned structure we are able to use the SVD to detect outputs in which the number of harmonic components has been reduced and thereby estimate the pitch of a note. We confirm the effectiveness of the proposed method by presenting the results of estimating pitch using the proposed method on polyphonic vocal music with both two and three voices singing at the same time.