Speech production acoustics in different language environment

在言语产生研究中，对发音-声学关系的了解是一个关键因素。嘴唇是声道的主要声学控制点之一，了解其几何特征与声学特征之间的关系至关重要。因此，对一组持续的元音和摩擦发音，制作了法国受试者嘴唇的录像和石膏复制品。复制品的声辐射阻抗用双麦克风方法测量，并用优化算法确定了单个均匀圆柱管的长度和面积。声学等效唇间面积与声道几何长度呈线性关系。采用三维有限元方法对声道简化模型声管的声学特性进行了数值模拟。结果表明，随着声壁阻抗分布面积的扩大或横截面变得更加平坦，共振峰频带宽度进一步增加，声波在共振峰频率附近的幅度减小。利用核磁共振技术测量了日语元音的声道形状，并利用有限元模型对复杂形状的声道三维效果进行了估计，在模型中引入了闭合段的伸长因子。我们指出了声道损失的影响，它只取决于横截面的周长。我们用机械激励的方法测量了声门开/关、高/低音等情况下的发声传递函数。我们看到，高频段对喉部的修饰敏感性较大。我们还利用我们的短时分析方法对实际发出的元音进行了分析，并讨论了上述影响。

In speech production research, a good knowledge of the articulatory-acoustic relationship is a key factor. As lips are one of the major vocal tract acoustic control points, it is of utmost importance to gain insight into relationship between their geometric and acoustic characteristics. Therefore, video recordings and plaster replica of the lips of a French subject were made for a set of sustained vowel and fricative articulations The acoustic radiation impedance of the replica was measured by a two-microphone method, and the length and area of a single uniform cylindrical tube were determined by a optimization algorithm. It has been shown that the acoustic equivalent interalabial area linearly depends on the geometric length.Acoustic characteristics of sound tubes as a simplified model of vocal tracts are simulated by a three dimensional Finite Element Method (FEM). The results suggested some new facts as follows ; if the area of the wall impedance distribution is spread or the cross section is flattened more, then the formant band widths are increase more, and the amplitude of sound wave is decreased near formant frequencies.The vocal tract shapes of Japanese vowels are measured by MRI.We estimated the 3D effect from the the complex shape by using our FEM models, in which we introduced an elongation factor for the closed section. We pointed out the effect of losses in the vocal tract which depends only on the circumference of the cross section. We measured vocal transfer functions by using mechanical excitation for some cases such as open/close glottis and high/low pitch. We saw that the sensitivity is larger in the high frequency region for the modification of the laryngeal part. We also analyzed real uttered vowels by using our short-time analysis, and discuss the above effects.