音声合成のモデル

Abstract

An overview of a speech synthesis system is given, with a special emphasis on hierachical structure to generate speech sounds from an input text. A speech synthesis system generally consisits of four phases: linguistic analysis of a text, phonological/phonetic processing, synthesis parameter generation, and sound synthesis. A statistical model to represent durations of phone segments and a linear time invariant system model to generate fundamental frequency contours are shown to have been successful in improving sound quality of synthetic speech sounds.