A parametric formulation of the generalized spectral subtraction method

Abstract

This thesis describes an acoustic noise suppression system for speech commu-nication applications when the noisy speech alone is available. Two short-time spectral amplitude estimators of the speech signal were derived and used for the system. The two estimators were derived based on a parametric formula-tion of the original generalized spectral subtraction method [6], [7], [21,pp. 509). The objective is to reduce the excessive level of residual noise and musical tone artifacts in the noise-suppressed speech commonly associated with the original method.