Studies on Glottal Source and Formant Trajectory Models for the Synthesis of High Quality Speech

Abstract

This chapter describes a polynomial voice source model and a formant trajectory model for a hi-fi speech synthesizer. The voice source model represents time derivative of the glottal volume velocity waveform as a polynomial function. The formant model describes the formant trajectories as the summation of temporal functions: a second order delay function, which represents vowel-to-vowel transitions, and two first order delay functions, which represent the effects of surrounding consonants on the vowel formant trajectories. The models were tested through perceptual experiments for synthetic speech at slow and fast speaking rates. Results suggest that the models work well particularly at slow rates. Some additional strategies seem to be needed to improve the intelligibility of consonants at fast rates.