Spectral methods for voice source parameters estimation

A spectral approach is proposed for voice source parameters representation and estimation. Parameter estimation is based on decomposition of the periodic and the aperiodic components of the speech signal, and on spectral modelling of the periodic component. The paper focusses on parameters estimation for the periodic component of the glottal ow. A new anticausal all-pole model of the glottal ow is derived. Glottal ow is seen as an anticausal 2-pole lter followed by a spectral tilt lter. The anticausal lter has complex poles, instead of the real poles that are usually assumed. Time-domain and frequency domain parameters are linked by analytic formulas. Two spectral domain algorithms are proposed for estimation of open quotient. The rst one is based on measurement of the rst harmonics, and the second one is based on spectral modelling. Experimental results demonstrate the accuracy of the estimation procedures. 1. INTRODUCTION In this paper we present our recent work on algorithms for automatic analysis of voice quality parameters. These parameters are studied in the spectral domain. Voice source measurements are needed for high quality speech synthesis, because voice quality is currently a key issue for naturalness. This is particularly true in the situation of speech synthesis using concatenation of natural speech segments (e.g. diphones or non uniforms units). Diierences in voice source, for instance diierences in vocal eeort, are often perceived by listeners as synthesis or concatenation errors, because of the change in quality across segments. Contrary to most of the recent works on source modelling, we prefered spectral processing, because it has a number of advantages: