A residual modeling extension of HNM model for prosodic modification of Slovak speech

Abstract

This article describes an extension of HNM model used for prosodic modification of speech. HNM model represents speech signals as a sum of harmonic and noise part. The decomposition of speech signal into these two parts allows more natural sounding modifications of the signal. The parametric representation of speech provides a straightforward way of changing prosodic features of the speech. Our algorithm, as an addition to HNM, models a residual (difference between original and HNM modeled signal) of the speech signal by method that allows us to modify the whole speech signal in better quality.