Multipitch estimation applied to single-channel audio source separation: relevant techniques and challenges

Abstract

The estimation of melody trajectories in single-channel polyphonic signals is a major field of study in digital signal processing, with principal applications in audio source separation, automatic music transcription and musical genre identi fication. The YIN algorithm and one multiple fundamental frequency estimator are tested using two different mixtures of harmonic sounds in order to identify advantages and limitations. A strategy is presented as a way to overcome these limitations and hence improve the accuracy of the estimated pitch trajectories.