Abstract
Most musical sounds show strong energy concentrations, such as regular time-frequency patterns of sonic partials, as a sparse energy distribution across the spectrographic analysis plane. Most studies of musical sound are based on strong spectral components such as the energy distribution patterns of the harmonic partials. However, between any two sonic partials most musical sounds show interesting low-energy signal patterns such as wide-band articulation or other performance noise. The authors present a musical signal analysis framework that extracts these low-energy spectrographic components by detecting and masking-off the strong sonic partials. A parametric estimation and analysis framework is implemented for an in-depth study of the timbral properties of these weak signal components. Based on the parametric models of several instrument categories, the authors implemented several resynthesis methods for these weak signal components.