Bayesian Data-Driven Subject-Specific Modeling of Voice Production

项目概要/摘要 该建议旨在发展语音中特定主题的贝叶斯发声计算模型 正常人和结构性嗓音障碍患者。发声是一个复杂的生物物理过程 过程，包括声带生物力学和声门下，声门内，声门上的空气动力学，作为 以及他们的互动。预测计算建模方法是非常需要的，因为它们提供了 科学工具，以更好地了解这样一个复杂的耦合系统的详细功能。他们可以 研究发声的正常功能，并调查它是如何受到影响的， 声带结构或行为的异常或故障。高速视频内窥镜的实验数据， 电声门图和声信号将用于设计声音产生的计算模型， 耦合喉部动力学和空气动力学。在目标1中，目标是开发贝叶斯预测模型 它可以捕捉数据和模型中固有的不确定性。将进行贝叶斯推断 使用高速视频内窥镜和电声门图数据。模型将通过声学验证 每个声音正常的参与者的信号。该模型将耦合声带组织振动（动力学和 运动学）与声门气流的瞬时相互作用的空气动力学，以考虑流动， 发声时的结构相互作用。在目标2中，目标是设计患者特定的计算模型， 结构性声音病变患者的发声，包括声带息肉、Reinke水肿和 喉炎假设是声带振动可以在患者中被强迫和流体诱导。一个 将根据患者的模型计算外部患者特异性分力，其中物理 声带的结构和振动行为受到病理学的负面影响。参数 将计算不确定性，并且预期由于疾病而在患者之间变化很大。的 这项研究的结果将扩展和加深我们对正常语音功能的理解， 嗓音障碍的病理生理学拟议的研究与所述的多个优先领域相协调 《2017 - 2021年国家发展战略规划》[3]。目标1支持优先事项1（"加深我们对 人类交流系统的正常功能"）， 生产规范。目标2提出确定发声的发声动力学和声门空气动力学 在结构性嗓音障碍患者中，这涉及优先事项2（"增加我们对 改变或削弱沟通和健康"）。这两个目标都支持优先事项3（"改进诊断方法， 治疗和预防"），通过确定哪些喉部机制被破坏的患者的声音 以及它如何影响声学信号。

Project Summary/Abstract This proposal aims to develop Bayesian subject-specific computational models of voice production in vocally normal individuals and patients with structural voice disorders. Voice production is a complex biophysical process, consisting of vocal fold biomechanics and sub-glottal, intra-glottal, and supra-glottal aerodynamics, as well as their interactions. Predictive computational modeling approaches are highly needed as they provide scientific tools for better understanding the detailed function of such a sophisticated coupled system. They can be employed to study the normal function of voice production and investigate how it can be impacted due to an anomaly or malfunction in the vocal fold structure or behavior. Experimental data of high-speed videoendoscopy, electroglottography and acoustic signals will be used to design computational models of voice production, coupling laryngeal dynamics and aerodynamics. In Aim 1, the objective is to develop Bayesian predictive models that can capture the uncertainties inherent in the data and models. The Bayesian inference will be performed using the high-speed videoendoscopy and electroglottography data. The models will be validated with acoustic signals for each vocally normal participant. The model will couple the vocal fold tissue vibration (kinetics and kinematics) with the instantaneously interacting aerodynamics of glottal airflow to take into account the flow- structure interaction during phonation. In Aim 2, the goal is to design patient-specific computational models of voice production for patients with structural voice pathologies including vocal polyps, Reinke's edema, and laryngitis. The assumption is that the vocal fold vibrations can be forced and fluid-induced in the patients. An external patient-specific force component will be calculated from the model for the patients, where the physical structure and vibratory behavior of the vocal folds are negatively impacted by the pathology. The parameter uncertainties will be calculated and expected to vary greatly among the patients due to the disorders. The outcome of this research will extend and deepen our understanding of the normal voice function and pathophysiology of voice disorders. The proposed research is in harmony with multiple priority areas described in the 2017-2021 Strategic Plan of the NIDCD [3]. Aim 1 supports Priority 1 (“deepen our understanding of the normal function of the systems of human communication”) by designing computational models of voice production for norm. Aim 2 proposes to determine vocal dynamics and glottal aerodynamics of voice production in patients with structural voice disorders, which addresses Priority 2 (“increase our knowledge about conditions that alter or diminish communication and health”). Both Aims support Priority 3 (“improve methods of diagnosis, treatment, and prevention”) through determining what laryngeal mechanisms are disrupted in patients with voice disorder and how it affects the acoustic signal.