Conclusions on analysis and synthesis of large semivocalic formantic transitions

Abstract

A very important issue in formantic Text-to-Speech (TtS) systems refers to formants transitions between various successive sounds in the word context. While the small-scale transitions can be treated as linear transitions, the large ones must be carefully considered in order to model the equivalent natural transition. When glottal excitation is active during such transitions, specific sounds are produced that require explicit treatment. Our approach to modeling large-scale formant transitions in Klatt-type synthesizers refers to dividing the large transition period into segments of reduced slope (relative steady state) and large slope (transitions between steady states). The short constant frequency segments are created by inserting the corresponding allophones, described by the respective formantic set. The modeling process is exemplified by the case of semivowel i. As a starting point for the synthesis, we analyze words containing semivowel i as part of diphthongs preceded by unvoiced or voiced palatal consonants. The speech synthesis of the analyzed words is depicted in terms of Klatt synthesis parameters.