Resolving Overlapping Harmonics for Monaural Musical Sound Separation using Fundamental Frequency and Common Amplitude Modulation

In mixtures of pitched sounds, the problem of overlapping harmonics poses a significant challenge to monaural musical sound separation systems. In this paper we present a new algorithm for sinusoidal parameter estimation of overlapping harmonics for pitched instruments. Our algorithm is based on the assumptions that harmonics of the same source have correlated amplitude envelopes and the phase change of harmonics can be accurately predicted from an instrument’s pitch. We exploit these two assumptions in a leastsquares estimation framework to resolve overlapping harmonics. This new algorithm is incorporated into a separation system and quantitative evaluation shows that the resulting system performs significantly better than an existing monaural music separation system for mixtures of harmonic instruments.