A parametric model for heart sounds

This paper presents a novel way to represent heart sound signals by using a harmonic plus noise model (HNM). A set of time-frequency waveforms are selected by the Matching Pursuit algorithm to form the harmonic part. The noise part is obtained after subtracting the harmonic part to the original signal and is modeled as an autorregresive process using linear predictive coding. The sum of both the harmonic and noise part models provides a highly accurate representation of the original heart sounds. The proposed model has many potential applications such as parametric low-bit rate coder, cardiac monitoring and heart disease identification.