In this paper we present and compare four time-domain approaches for estimating the parameters of a harmonic speech model. The classic approach of Least Squares is directly compared with a Total Least Squares approach trying to overcome errors in the estimation of the fundamental frequency of the model. Both of these approaches are suboptimal since they split the estimation problem into two subproblems; to the estimation of amplitudes and phases and to the estimation of fundamental frequency. To improve the accuracy of the parameters estimation of the harmonic model two iterative non linear approaches are then presented, based on the Steepest Descent and Newton-Gauss optimization algorithms, where all parameters of the harmonic model are estimated simultaneously. The approach based on the Newton-Gauss optimization algorithm provided the best accuracy as this is measured by the Signal-to-Noise Ratio criterion.
Cite as: Pantazis, Y., Rosec, O., Stylianou, Y. (2008) On the estimation of the speech harmonic model. Proc. ISCA ITRW on Speech Analysis and Processing for Knowledge Discovery, paper 031
@inproceedings{pantazis08_spkd,
author={Yannis Pantazis and Olivier Rosec and Yannis Stylianou},
title={{On the estimation of the speech harmonic model}},
year=2008,
booktitle={Proc. ISCA ITRW on Speech Analysis and Processing for Knowledge Discovery},
pages={paper 031}
}