New distortion measures for speech processing

Abstract

New distortion measures are derived from a recently proposed characterization function of stationary time series and are shown to be more robust than some commonly-used distortion measures such as the Kullback-Leibler spectral divergence in speech processing.