A Comprehensive Psychoacoustic Approach to Voice Quality Perception

项目总结/摘要 语音障碍往往会导致语音质量的变化，患者，临床医生和谈话注意到 伙伴语音质量的评估对于诊断是必不可少的，并且语音质量的改善是关键的。 治疗的结果。然而，我们对语音质量如何感知的知识是有限的， 对临床结果的可靠测量更加有限。这限制了我们准确地 量化或描述质量的变化，例如由于疾病导致的下降或由于 治疗这个延续项目结合了声音科学，语音科学， 听科学和工程来解决这些问题。全面处理拟议的 研究同时涵盖了三个主要的VQ维度，包括他们的协方差，提高 测量方法，并通过连接语音评估扩展生态效度。研究 首先在实验室中获得语音质量感知的高精度测量。这些数据 然后用于开发准确反映听众数据的语音质量感知的数学模型。 为了获得人类对语音质量的判断与模型输出之间的紧密匹配，听觉模型 处理用于获得语音声学信号的内部表示。具体措施 然后从该内部听觉表示中捕获并用于对语音质量的感知进行建模。在 建议的工作，这些方法将用于建立一个全面的框架， 测量语音质量感知并能够转化为临床实践。具体目标是 项目是：（1）发展一个更完整的理解VQ协方差的性质（在呼吸， 粗糙和应变）使用自然的dysphonic声音;（2）评估VQ在连接语音在微观（分段） 和宏观（整个话语）水平，以更好地捕捉快速和复杂的过渡，共同的影响， 发音和韵律变化，以及与特定病理相关的独特VQ签名;（3）将 所有这些组件到一个新的，三维VQ缩放过程，是简单的，直观的，有效的 足以在临床上使用，并具有比率级测量特性;（4）使用高度预测性 计算模型，以克服声学分析和感知评估的许多限制;以及 (5)评估测试设计和评估中考虑的全方位心理测量特性，包括可靠性， 有效性、灵敏度和特异性。也许最重要的是，目标是制定一项评估， 准确地指示对变化的响应（即，疾病进展、治疗）。这些参数 被认为是纵向评估和结果衡量的关键。使用这些的可行性 定期临床评估中的模型和指标将在多种语音病理中通过 临床现场试验和临床医生和实验室受试者的协调离线评估。

PROJECT SUMMARY/ABSTRACT Voice disorders often lead to changes in voice quality noticed by patients, clinicians, and conversation partners. Assessment of voice quality is essential for diagnosis, and improvement in voice quality is a critical outcome of treatment. However, our knowledge of how voice quality is perceived is limited and the availability of robust measures of clinical outcome is even more limited. This has restricted our ability to accurately quantify or describe changes in quality, such as declines due to a disease or improvements resulting from treatment. This continuation project combines concepts and techniques from voice science, speech science, hearing science, and engineering to address these problems. The comprehensive approach to the proposed research simultaneously envelops the three primary VQ dimensions, embraces their covariance, improves measurement methods, and expands ecological validity through connected speech evaluation. The research proceeds by first obtaining high-precision measures of voice quality perception in the laboratory. These data are then used to develop mathematical models of voice quality perception that accurately reflect listeners’ data. To obtain a close match between human judgments of voice quality and model output, models of auditory processing are used to obtain an internal representation of the voice acoustic signal. Specific measures are then captured from this internal auditory representation and used to model the perception of voice quality. In the proposed work, these methods will be used to establish a comprehensive framework for understanding and measuring voice quality perception and to enable translation to clinical practice. The specific goals of this project are to: (1) develop a more complete understanding of the nature of VQ covariance (among breathy, rough, and strain) using natural dysphonic voices; (2) assess VQ in connected speech at micro (segmental) and macro (whole utterance) levels to better capture the impacts of rapid and complex transitions, co- articulatory and prosodic variations, and unique VQ signatures related to specific pathologies; (3) incorporate all of these components into a novel, three-dimensional VQ scaling procedure that is simple, intuitive, efficient enough to be used clinically, and has ratio-level measurement properties; (4) use highly-predictive computational models to overcome many of the limitations of acoustic analyses and perceptual evaluation; and (5) evaluate full range of psychometric properties considered in test design and evaluation including reliability, validity, sensitivity, and specificity. Perhaps most importantly, the goal is to develop an assessment that accurately indicates responsiveness to change (i.e., disorder progression, treatment). These parameters are considered essential for longitudinal assessment and outcome measurement. The feasibility of using these models and metrics in regular clinical assessment will be evaluated in multiple voice pathologies through a clinical field-trial and coordinated offline assessments by clinicians and laboratory subjects.