Natural Sound Speech Synthesis by 3D Acoustic Model Based on MRI Measurement of Speech Production System

本研究的目的是详细了解鼻腔细微形态结构的声学效果。利用磁共振成像技术（MRI）对鼻腔进行测量，在计算机上建立了几种不同的数值声管模型。利用有限元法计算了模拟鼻道的声压、质点速度、声强和声学传递函数，模拟鼻道实质上是一个从软腭到鼻孔的类似于人鼻道的声管，其间有左右鼻道分叉和几对鼻窦。将椭圆形模型做成截面为椭圆形的直管，其面积和截面积分别与实际尺寸相同。制作了一个与实际鼻道相似的弯曲、复杂截面的真实模型。上颌窦和上颌窦 ...更多信息 额窦，也介绍了。此外，还研究了鼻道横截面变窄和在鼻道内某处插入障碍物的声学效应，结果表明，鼻道的面积函数不仅代表了鼻道沿着轴向的横截面面积，而且还反映了鼻道的横截面形状、分叉、左右不对称、鼻道的弯曲也影响声学特性。我们证明了直椭圆管模型近似能力的极限。鼻道的左右不对称性使传递函数产生零极点对。鼻道的复杂形状和弯曲通过鼻腔中的环状声强流产生传递函数的零点。证实了窦腔本身作为亥姆霍兹谐振器，在传递函数中在其谐振频率附近产生一个零极点对。然而，实际上，在鼻腔的两侧存在多个窦腔，并且它们是不对称的，因此，它们在2000Hz以下产生相当复杂的声学特性。一侧鼻道狭窄导致鼻道左右不对称性加重，使鼻道的传输性状复杂化。最后，它表明，任何障碍物插入鼻道中可能会产生一个分支流的声波在鼻道中，并可能会导致一些极点零的传递函数。少

This research aims to find in detail the acoustic effects of fine morphological construction of the nasal cavity. Several kinds of numerical acoustic tube models were constructed on a computer from measurements of the nasal cavity by Magnetic Resonance Imaging technology (MRI). Using FEM (finite element method), we calculated acoustic pressure, particle velocity, acoustic intensity, and acoustic transfer function of the simulated nasal tracts.The simulated nasal tract is essentially an acoustic tube which is similar in some aspect to human nasal tract from the soft palate to the nostrils, between which there are a bifurcation of left and right nasal passages and several pairs of the sinuses. A elliptic model was made as to be straight tube having elliptic cross section in which an areas and coircumference were identical with the actual dimensions, respectively. A realistic model was made as to be bent and complicated in cross section as in actual nasal tract. The maxillary sinus and th … More e frontal sinus, were also introduced. In addition, the acoustic effcts of narrowing the nasal tract cross section, and inserting something obstacles at somewhere in the nasal tract were examined.From the results, it is found that not only the conventional area function of the nasal tract which represents cross sectional areas of the nasal tract along with axial locations, but the cross-section shape, the bifurcation, the left-right asymmetry, and the bending of the nasal tract also affect the acoustic characteristics. We showed a limit of approximation ability of straight-elliptic tube models. The left-right asymmetry of the nasal tract always produce pole-zero pairs in transfer function. The complicated shape and the bending of the nasal tract produces zeros of the transfer function by a circle like flow of acoustic intensity in the nasal cavity. It is confirmed that the sinus cavity itself acts as a Helmholtz resonator and produces a pole-zero pair at nearby its resonance frequency in the transfer function. However, there are actually several number of sinus cavities in both sides of the nasal tract, and they are not symmetrical, therefore, they produce fairly complex acoustic property below 2000Hz. The narrowing of one side of the nasal tract causes a severer left-right asymmetry and produces a complexity in transfer characters. Finally, it is shown that any obstacle inserted in the nasal tract may produce a branching flow of acoustic wave in the nasal tract and may cause some pole-zero in the transfer function. Less