Separation of harmonic sounds using multipitch analysis and linear models for the overtone series

Abstract

A signal processing method for the separation of concurrent harmonic sounds is described. The method is based on a two-stage approach. First, a multiple fundamental frequency estimator is applied to find initial sound parameters which are reliable, but inaccurate and static. Second, time-varying sinusoidal parameters are estimated in an iterative algorithm. The harmonic structure is retained by keeping the frequency ratio of overtones constant over time. Overlapping harmonic components are resolved using linear models for the overtone series. In practice, the models retain the spectral envelope continuity of natural sounds. Simulation experiments were carried out using generated test signals, which were random mixtures of two to six notes from recorded natural instruments. The system is able to produce meaningful results in all polyphonies, the quality of separated sounds gradually degrading along with the polyphony. Some denoising algorithms were applied to suppress nonstationary noise component, such as drums in real-world music signals. However, the usability of the system for real musical signals is still quite limited.