Articulatory model based on articulatory data and speech synthesis

本研究的目的在于发展一个模拟人类语音产生机制的语音合成系统。为了产生包括连续语音在内的语音，研制了一种模拟语音传输过程中声道传递函数的声道模拟器。从磁共振（MR）图像中获得稳定元音产生过程中的声道形状。利用级联一阶系统的函数模型模拟了连续语音中声道形状的变化。该模型能够以较高的精度描述发音轨迹。由于声道模拟器能够平滑地描述声道形状的变化，因此获得了自然的语音输出。本研究亦利用有限元素法分析，探讨级联声管对声道形状的近似对共振峰频率的影响。在此基础上，开发了基于图形用户界面（GUI）的语音合成与分析工具，为进一步研究语音生成过程提供了基础。

This study aims at advancement of the speech synthesis system which simulates a mechanism of human speech production process. A vocal tract simulator, which simulates transfer function of vocal tracts during speech, was made in order to produce speech sound including continuous speech. Vocal tract shapes during production of steady vowels were obtained from magnetic resonance(MR) images. The transition of vocal tract shapes during continuous speech was simulated by means of a functional model based on cascaded first-order systems. The model can describe articulatory trajectories with a high accuracy. Natural speech output was obtained because the vocal tract simulator was able to describe smoothly the change in vocal tract shapes. Effects of approximation of the vocal tract shape by cascaded acoustic tubes on formant frequencies were also investigated by means of finite element method (FEM) analysis in this study. Furthermore, based on these results, graphical user interface (GUI) based speech synthesis and analysis tool was developed for further study of speech production process.