基于听音者面部表情分析的噪声烦恼度评估新方法

紧扣噪声烦恼度主观测量的应用短板，本项目提出一种基于情绪表情分析的噪声烦恼度评估新方法，即利用听音者面部表情开展噪声情景现状分析与烦恼度评估。该方法具有实时灵敏、结果直观、绿色低成本等显著优势。首先，采用时间轴分析法对听音者因受到噪声干扰而烦恼时的面部表情进行特征区块编码、主要动作单元运动模式的分级描述研究；创建面部表情图示知觉模型，为噪声情景标注情感标签；提出基于面部表情的听觉情感分析法，分级评价噪声烦恼度。随后，整合声特性（声）、听音者烦恼度（情）与面部表情（行）三方线索，建构三项闭环模型，用于声环境状况在线分析与听音者情绪实时预测，为实现非传感器环境声监测和全域移动环境声感知奠定基础。本研究旨在突破现阶段烦恼度评估的技术瓶颈，积累噪声心理效应研究新经验，开拓声环境质量评价与控制方法研究新视野，为创建集声环境监测、控制与美化设计于一体的智能平台规划新路径，引领“智慧环保”新范式。

To clasp the short-board in practical application of noise-related annoyance (NoRA) assessment by subjective measurements, an original strategy use facial expression analysis rather than the verbal/number scale to estimate NoRA levels and depict acoustic scenarios is proposed in our project. Our fresh method has the advantages of real-time sensitivity, intuitive results, green and low cost. Firstly, the time axis analysis technique is used to analyze the facial expression of the listener who annoyed by noise. Then the motion patterns of the main facial action units are classified and depicted. A perceptual graph model based on facial expressions is created to label the noise scene and to calculate the auditory emotional states. Subsequently, a three polynomial closed-loop model would be proposed by integrating clues extracted from the noise features, the self-reported annoyed levels and the facial expressions, to achieve the on-line analysis of acoustic scenarios and real-time prediction of the listener’s emotional states, which laid the foundation for non-sensor monitoring and global dynamic perception of environmental sound quality. Our efforts aim to break through the technical bottleneck of annoyance assessment, to explore new experience in noise psychological effects, and to expand a new perspective to environmental sound quality evaluation besides soundscape design. The ultimate goal is to draw new path for building an artificial intelligence platform setting online noise monitoring, controlling and improving in one, to lead an innovation paradigm of intelligent environment.