Acoustic of Vocal Tract Shapes for Liquids

描述(由申请人提供)：有重要的个人 人们产生语言音素的方式的差异。非常 声道结构中不同但系统的变化可能会产生 母语人士认为绝对等同的声学特征。这个 发音复杂的美国英语液体/r/和/L/，尤其是， 展示几种不同类型的舌形。值得注意的是，这些 音素对儿童和非母语人士来说是最麻烦的， 但目前的声道声学模型并没有考虑到个体 在生物反馈语言治疗中可能有用的方式的变化。为 例如，治疗师目前不能“读取”光谱信息来发现 在声道的几种狭窄中，哪一种导致了失真/r/。 此外，通过超声或腭部造影术进行的发音反馈仅显示 声道的一部分。以前的声学模型由于缺乏 3维数据。利用核磁共振技术，我们提出了获取多说话人数据库的建议 声道的体积和形状。这些体积数据将用于显影 声道说话人差异的综合声学模型 配置。我们的目标是根据声学对声道形状进行分类 制作/r/或/1/的策略。我们还将获取一部分超声数据 舌头在同一组扬声器上的纵向轮廓。到时候我们会的 测试超声曲线部分约束分类的能力 全长MM中矢状舌轮廓形状到预先定义的 配置类型。这些数据将有助于我们对声道的理解 语音中的声学和发音变化。它们可能被用来改进 语音识别技术及其在语音识别系统中的实现 声学生物反馈治疗技术。

DESCRIPTION (provided by applicant): There are significant individual differences in the way people produce the phonemes of their language. Very different but systematic variations in vocal tract configuration may produce acoustic profiles that native speakers accept as categorically equivalent. The articulatorily complex American English liquids /r/ and /l/, in particular, show several distinct types of tongue configuration. Significantly, these phonemes are among the most troublesome for children and non-native speakers, but current models of vocal tract acoustics do not account for individual variations in ways that might be useful in biofeedback speech therapy. For instance, therapists cannot currently "read" spectral information to discover which of several constrictions in the vocal tract is causing a distorted /r/. Further, articulatory feedback via ultrasound or palatography shows only a portion of the vocal tract. Previous acoustic models were limited by the lack of 3-dimensional data. Using MRI, we propose to acquire a multispeaker database of vocal tract volume and shape. These volumetric data will be used to develop comprehensive acoustic models of interspeaker differences in vocal tract configuration. Our goal is to categorize vocal tract shapes in terms of acoustic strategies for producing /r/ or /1/. We will also acquire ultrasound data for a portion of the longitudinal profile of the tongue on the same set of speakers. We will then test the ability of ultrasound curve portions to constrain categorization of full-length MM midsagittal tongue profile shapes into previously defined configuration types. These data will aid our understanding of vocal tract acoustics and articulatory variation in speech. They may be used to improve speech recognition technologies and the implementation of articulatory and acoustic biofeedback therapy techniques.