The role of anatomical factors in speech production: towards the speaker-independent articulatory model.

本研究的总体目标是提高我们对言语发音控制的说话人间差异的理解。除了理论上的重要性外，这一知识还具有许多重要的实际意义。其中包括开发一个独立于说话人的语音识别和合成模型，以及创建一个全面的语音运动参考数据库，该数据库可用于跨语言和语音生成训练研究。***这个应用程序的具体重点是在语音生产变异性的解剖学因素。这条调查线的动机是由于文献中缺乏关于诸如口腔和舌头大小，头骨形态和颌骨长度与言语和口腔运动特征之间存在关系的协议。在这项研究中，我们将探讨以下问题：(1)能否根据说话人的口腔解剖测量来分类？(2)具有相似解剖结构的说话者在讲话中是否表现出相似的发音运动模式？(3)解剖变异是否导致了说话时舌颌协调和发音策略的差异。***将使用计算机辅助侧位头测术获得60名成年人的解剖数据。大量不同复杂程度的语料库（例如，音节到短语到段落），风格（例如，快，慢，清晰）和非言语任务（例如，最大张嘴）将使用电磁发音仪记录。解剖学测量（例如，舌头大小，下颌和下颌骨的大小和形状）将用于聚类分析，以将说话者分类。运动学测量将包括那些专注于发音运动作为一个整体（例如，发音工作空间，运动动态范围），声音特异性（例如，发音位置可变性）和音节/短语特异性测量（例如，平均移动速度，峰值速度，移动持续时间和范围）。它们将与解剖测量相关联，并在不同解剖组的说话者之间进行比较。***这项研究将是第一个全面研究发音器解剖尺寸的变异性及其在大量说话者的言语运动变异性中的作用的研究。这项工作的结果将导致识别在说话者之间共同的语音产生中的“规范”发音参数，并促进独立于说话者的语音产生发音模型的发展。

The overall goal of this research is to improve our understanding of inter-speaker variability in articulatory control of speech. Along with the theoretical importance, this knowledge has a number of important practical implications. These include the development of a speaker-independent model for speech recognition and synthesis and creation of a comprehensive speech movement reference database that can be used in cross-linguistic and speech production training studies. *** The specific focus of this application is on the anatomical factors underlying speech production variability. This line of enquiry is motivated by the lack of agreement in the literature on the existence of a relationship between factors such as oral cavity and tongue size, skull morphology and jaw length with speech and oral movement characteristics. In this study we will pursue the following questions: (1) Can a speaker be classified based on measures of their oral anatomy? (2) Do speakers with similar anatomies show similar articulatory movement patterns in speech? and (3) Does anatomic variability contribute to differences in tongue-jaw coordination and articulatory strategies during speaking. *** Anatomic data will be obtained from 60 adults using computer-aided lateral cephalometry. A large corpus of speech of different complexity (e.g., syllable to phrase to paragraph), style (e.g., fast, slow, clear) and non-speech tasks (e.g., maximum mouth opening) will be recorded using electromagnetic articulography. Anatomic measurements (e.g., tongue size, jaw and mandible size and shape) will be used in a cluster analysis to classify speakers into groups. Kinematic measures will include those focused on articulatory movements as a whole (e.g., articulatory working space, motion dynamic range), sound specific (e.g., articulatory position variability), and syllable/phrase specific measures (e.g., average movement speed, peak speed, movement duration, and range). They will be correlated to the anatomic measurements and compared between speakers in different anatomic groups. *** This research will be among first to comprehensively examine the variability in the anatomical dimensions of articulators and their role in speech movement variability in a large cohort of speakers. The outcome of this work will lead to identification of “canonical” articulatory parameters in speech production common across speakers and facilitate the development of the speaker-independent articulatory model of speech production.