精神压力下基于物理模型的变异语音生成机理探索及检测方法研究

The speech under stress induced by high workload, multi-tasking, fatigue, and emotional situations has been the subject of many studies. Methods for stress classification used at present have their limitation because of the lack of a description for the stressed speech production. In this project, the physical model for stressed speech production is proposed, and the production mechanism for the stressed speech is studied by focusing on variations in the aerodynamics of airflow patterns in vocal system physiology. The relationship of physiological feature related to stress and parameters of glottal flow is established by physical modeling, and the algorithm is designed to extract the glottal-flow parameters with physical meaning. Also an objective evaluation standard for stress based on physiological signals is proposed to label the speech samples during the data collection process. Main focus: (1) Physical modeling for the stressed speech. (2) Algorithm design for feature extraction based on physical model for stress classification. (3) An objective evaluation using physiological signals. The study based on speech production could process verbal and non-verbal information in speech signal, and lay the theoretical and practical foundation for related area of speech technology.

压力、情感、心理紧张引起的说话人发声变异一直是语音领域的研究热点。因变异语音生成机理复杂，且缺少统一的表征和描述，使得精神压力下语音的检测识别存在着一定局限性。为此，本研究基于传统模型，建立说话人变异语音生成的物理模型，揭示精神压力状态下说话人发声生理系统及其中气流流态的变化规律，探索变异语音的生成机理；通过物理模型模拟，设计生理特征估计算法，建立压力相关生理特征与声门波参数的关联，提取与压力因素有关且拥有物理意义的声门波参数；以说话人生理信号作为压力异常状态的客观评价标准，在语音收集的过程中对样本进行标注。重点解决：（1）建立针对变异语音的物理生成模型；（2）设计基于物理模型的特征提取算法，进行压力下变异语音识别；（3）通过说话人生理信号进行压力的客观评价。研究成果不仅可以处理语音中的言语和非言语信息，而且可为语音技术的相关研究领域问题的解决奠定了理论和实践基础。

本项目基于传统模型建立变异语音生成的物理模型，揭示压力状态下发声生理系统工作运动机制以及其中气流流态所受的影响和变化规律，对变异语音的内在生成机理进行探索性研究。基于模型估计说话人生理特征，阐明生理特征与声门波参数之间的内在联系，建立基于生理信号的压力客观评价标准，设计变异语音的识别检测算法。为该技术在实际中的应用打下扎实的理论基础和实验基础。