Pitch tracking in polyphonic audio by clustering local fundamental frequency estimates

Abstract

A novel way of performing pitch tracking by means of clustering local fundamental frequency (f0) candidates is described. The technique is based on an existing pitch salience representation for polyphonic music called F0gram which relies on the Fan Chirp Transform. The grouping is performed by applying a Spectral Clustering method, since it can handle filiform shapes such as pitch contours. The approach seems appealing since many sound sources can be tracked simultaneously and the number of contours and sources is derived from the data. Results of a melody detection evaluation indicate the introduced method is promising, despite that various aspects of the technique deserve further work.