Variability of vowel formants due to the finite impedances of the glottis, the velopharyngeal port, and the vocal tract walls

元音共振峰频率通常被认为主要由声道的腔体形状决定，其带宽由各种损耗机制决定，例如软壁，粘性摩擦和嘴唇的声音辐射。然而，一些研究表明共振峰频率也受到某些边界条件的影响，特别是发声过程中声门的有限声阻抗，声道的软壁以及不完全的咽喉闭合。所有三种机制的效果都是相似的：前两个共振峰被转移到更高的频率，它们的带宽变得更大（与具有刚性壁的声道相比，声门完全关闭，以及完全关闭的咽喉口）。到目前为止，这三种机制只被单独考虑，大多数情况下使用语音装置的物理模型或使用计算机模拟（而不是人类）。本文提出的项目的目标是通过结合受试者实验、物理声道模型和计算机模拟来更详细地研究这些机制。特别感兴趣的是多个并发机制如何相互作用，在转移共振峰，他们共同的物理原因是什么，在多大程度上的影响是感知相关的，以及他们是否和如何故意使用在讲话或唱歌。该项目的结果将有助于更好地解释所测得的共振峰频率，改进从共振峰频率估计声道长度或声道形状的方法，改进逆滤波方法，以及更准确的语音产生模拟模型。

Vowel formant frequencies are often thought to be determined primarily by the cavity shape of the vocal tract, and their bandwidths by various loss mechanisms such as the soft walls, viscous friction, and sound radiation from the lips. However, several studies suggest that formant frequencies are also considerably affected by certain boundary conditions, in particular by the finite acoustic impedance of the glottis during phonation, by the soft walls of the vocal tract, and by incomplete velopharyngeal closure. The effect of all three mechanisms is similar: the first two formants are shifted to a higher frequency and their bandwidths become larger (compared to a vocal tract with rigid walls, a fully closed glottis, and a fully closed velopharyngeal port). So far, the three mechanisms have only been considered individually, and mostly using physical models of the speech apparatus or using computer simulations (rather than humans). The goal of the project proposed here is to study these mechanisms in more detail by combining experiments on subjects, with physical vocal tract models, and computer simulations. Of particular interest is how multiple concurrent mechanisms interact in shifting the formants, what their common physical cause is, to what extent the effects are perceptually relevant, and whether and how they are intentionally used in speech or singing. The results of the project would allow, among other things, a better interpretation of measured formant frequencies, improvement of methods for estimating vocal tract length or vocal tract shape from formant frequencies, improvement of inverse filtering methods, and more accurate simulation models of speech production.