基于人耳生理结构的车内噪声声品质预测  

Sound quality prediction of vehicle interior noise based on physiological structure of human ear

作　　者：刘兆海 张波 贺志恒 赵禹 刘后广[1] Liu Zhaohai;Zhang Bo;He Zhiheng;Zhao Yu;Liu Houguang(School of Mechanical and Electrical Engineering,China University of Mining and Technology,Xuzhou 221116,China;Guangzhou GRC Metrology&Test Co.,Ltd,Guangzhou 510600,China;Auto Engineering Research Institute,BYD Auto Industry Co.,Ltd,Shenzhen 518118,China)

机构地区：[1]中国矿业大学机电工程学院,徐州221116 [2]广州广电计量检测股份有限公司,广州510600 [3]比亚迪汽车工业有限公司汽车工程研究院,深圳518118

出　　处：《仪器仪表学报》2024年第12期284-294,共11页Chinese Journal of Scientific Instrument

基　　金：国家自然科学基金(52275296);江苏省研究生科研与实践创新计划(KYCX24_2720);中国矿业大学研究生创新计划(2024WLKXJ076)项目资助。

摘　　要：针对现有车内噪声声品质预测模型无法分析不同听觉状态人群的噪声感知差异的不足,提出了一个基于真实人耳生理结构的车内噪声声品质预测模型。首先,采集3款轿车的车内噪声样本,并通过主观评价实验得到噪声的主观评价值。然后,通过结合外耳滤波器模型、中耳集总参数模型、耳蜗集总参数模型和听毛细胞纤毛流体耦合模型,构建了基于人耳生理结构的听觉外周模块;以泄露积分激发神经元模型模拟听神经兴奋,利用深度神经网络(DNN)模拟听觉中枢对声音的感知,构建出生成生理响度、生理尖锐度与生理粗糙度的仿听觉中枢模块;基于仿听觉中枢模块生成的心理声学参数,通过TabNet模型得到声品质,构建声品质决策模块。听觉外周模块、仿听觉中枢模块和声品质决策模块构成声品预测模型。最后,对比分析该声品质预测模型与目前声品质模型的预测结果。研究结果表明,所提声品质预测模型能够较好地预测出车内噪声的声品质,该模型预测结果与主观评价结果的平均误差为3.3%,低于采用人工神经网络决策的6.4%和采用Zwicker模型计算心理声学参数的7.7%。该模型为研究不同听觉状态人群的车内噪声声品质提供一种新方案。A new sound quality prediction model for vehicle interior noise,based on the physiological structure of the human ear,is proposed to address the limitations of existing models that fail to effectively analyze differences in interior noise perception among individuals with varying hearing states.The model begins with the collection of interior noise samples from three cars,followed by subjective evaluation experiments to obtain subjective noise ratings.An auditory peripheral module is then constructed,incorporating an outer ear filter model,a middle ear lumped parameter model,a cochlear lumped parameter model,and a hair cell ciliary fluid coupling model to simulate the physiological structure of the human ear.An imitative auditory center module,designed to generate physiological loudness,sharpness,and roughness,is built using the Leakage Integral-and-Fire neuron model to simulate auditory nerve excitation,and deep neural networks to replicate the auditory center’s sound perception.The sound quality decision module is developed by integrating the psychoacoustic parameters generated by the auditory center module into a TabNet model to predict sound quality.Together,the auditory peripheral module,the imitative auditory center module,and the sound quality decision module form the complete sound quality prediction model.Finally,the model′s predictions are compared with those of existing models.Experimental results demonstrate that the proposed model accurately predicts vehicle interior noise quality,with an average prediction error of just 3.3%,outperforming the 6.4%error of artificial neural network-based models and the 7.7%error using the Zwicker model for psychoacoustic parameter calculation.This model offers a novel approach for studying the sound quality of in-vehicle noise for individuals with varying hearing states.Since the existing quality prediction model cannot effectively analyze the differences in interior noise perception of people with different hearing states,a sound quality prediction model of vehi

关 键 词：车内噪声 声品质 心理声学 人耳生理结构 深度学习 

分 类 号：TH701[机械工程—仪器科学与技术] TB533.2[机械工程—精密仪器及机械]