Tracking of partials in music signals using Kalman filtering: modeling and analysis

In this paper we propose a method for tracking of partials in music signals using Kalman filtering. Our observations are frequency and amplitude of peaks from spectral representation of short segments of music signal. These peaks are detected using a novel technique. We also introduce a set of state-space models for evolution of partials in time. Parameters of these models can be estimated by statistical analysis of a large database of musical sounds. Since Kalman filter is sensitive to the accuracy of these parameters, we propose some robust filters that can improve the performance of our tracker and discuss the possibility of using them.