声道模擬による音声の合成について

概要

論文の詳細を見る
This paper describes a model for speech synthesis by articulatory movements controlled by modified time optimal control. That is, the lateral shape of the vocal tract for each phoneme is decomposed into two component shapes for the tongue and the jaw. Then each point on each articulator is moved toward the corresponding point on the target shape under the assumption of a modified time optimal control. The control has two models : vowel type and consonant type, and the control forces are automatically adjusted considering the features of each phoneme and each articulator, the distance between the initial and target shapes, and the speed of utterance. Thus the model can simulate the processes of coarticulation at an arbitrary speech rate and the duration of each phoneme is decided naturally. Two target shapes C' and C'' for a consonant /C/ are prepared, and each of them is used in connection with either a front vowel or a back vowel. For /C/ in /V_1 CV_2/ (V : vowel), a concatenation C' C'' should be used relating to the first and last vowels.
山梨大学の論文