A robust algorithm for partial tracking of music signals
In this paper we propose a novel approach for tracking of partials in music signals based on a robust Kalman filter. Our tracker is based on a regularized least-squares approach that is designed to minimize the worst-possible regularized residual norm over the class of admissible uncertainties at each iteration. We introduce a set of state-space models for our signals based on the evolution of frequency and amplitude in different classes of musical instruments. These prior models are used to estimate future values of partial tracks in successive time frames of our spectral data. Here, the parameters of evolution models are treated as bounded uncertainties and our tracker can robustly track partials in all frequency regions. Unlike the conventional Kalman tracker, performance of this tracker is not influenced by the magnified track variations in higher frequencies. This tracker promises an improved performance over conventional Kalman tracker while preserving its good properties and superiority over existing methodologies.