Recent advances in Japanese speech synthesis research

Abstract

This paper introduces new trends in Japanese speech synthesis research focussing on computational modeling of control rules and their statistical optimization. To be free from time-consuming system building efforts, unit selection synthesis techniques have been proposed and intensive studies have been carried out on unit selection criteria and measures, as well as on selection algorithms and speech corpora from which units are extracted. As for prosody control, statistical optimization has been carried out using large speech corpora. These research efforts have enabled intelligible and very natural synthetic speech.< >