基于降维-回归的车内声品质时变烦躁度评价

Evaluation of Time Varying Annoyance for Vehicle′s Interior Sound Quality Based on Dimension Reduction and Regression

用心理声学参量表征声品质评价指标,可直观描述人对噪声信号的主观感觉。相比于听审团评分,它具有简单快速、效率高的优点,在此基础上将多个客观参量降维到低维空间,用少数具有代表性的参量定量表征主观感受,有利于声品质评价的标准化。通过采集轿车车内噪声信号,以等间隔不同车速状态和怠速状态下的噪声样本作为研究对象,考虑汽车车内噪声特性的时变效应,分别计算出噪声样本的心理声学客观参数时变算术平均值。提出以“时变烦躁度”作为主观评价指标,以语义细分法结合数字等级评分法作为主观评价方法进行主观评价试验。用因子分析对6种客观参量进行降维,结合显著相关性分析确定出响度(Z)、粗糙度、AI指数3个客观参量,并用多元回归建立声品质烦躁度评价预测数学模型。预测结果显示:回归预测三参量模型的相对误差率在5%以下,比六参量模型相对误差率低,预测效果比较理想,说明这3个参量可以有效表征主观烦躁度,同时证明噪声声品质的时变烦躁度评价预测模型是有效的。

Using psychoacoustic parameters to characterize the sound quality evaluation index can directly describe people′s subjective feeling of noise signals.Compared with the audio panel scoring,this method has the advantages of simple,fast and efficient.In this method,the objective parameters are reduced to a low dimensional space.Only using a small number of representative parameters can quantitatively describe subjective feeling,which is favorable to the standardization of sound quality evaluation.In this paper,the interior noise signals of a car is collected,the noise samples at equal spacing points of different constant speed and idle speed are taken as the research objects.Considering the time varying effect of the interior noise characteristics of the car,the time varying arithmetic mean values of the objective psychoacoustic parameters of the noise samples are calculated.“Time varying annoyance”is proposed and used as the subjective evaluation index.And the semantic differential method combined with grade scoring method is used for the subjective evaluation.Factor analysis is used to reduce the dimension of 6 objective parameters.Combined with significant correlation analysis,three objective parameters,such as loudness(Z),roughness and AI index,are selected.And the mathematical model of sound quality evaluation and prediction is established by multiple regression.The results show that the relatively error rate of the three-parameter model of regression prediction is below 5%,which is lower than that of six-parameter model,and the prediction effect is ideal.So,it shows that these three parameters can effectively characterize subjective annoyance,and proves that the proposed time varying annoyance evaluation prediction model of noise sound quality is effective.