摘 要

汽车行业作为我国经济的支柱型产业发展迅速，目前我国已经是世界汽车生产第一大国。汽车工业的发展以及生活水平的提高使得人们不仅关心汽车的实用性，而且对影响汽车舒适性的汽车噪声也变得越来越关注了。通过阅读大量的国内外在心理声学研究领域的文献后发现，学术界目前研究汽车稳态噪声的主客观评价的文献非常少，而这对于改善汽车噪声污染问题又是非常有必要的。

首先，对汽车的进气系统结构以及进气噪声的产生原理进行了分析，充分地了解了汽车进气噪声的相关知识。接着进行了汽车进气噪声的测量实验，并取得了原始的进气噪声数据信号。通过matlab的自编软件以及LMS系统的后处理软件将实验所得的噪声信号进行筛选和后处理，将处理过的声音文件进行国际标准化，最终建立了本次研究所用的声音样本库。

其次，对心理声学方面的知识做了详细的介绍，包括对人体听觉系统以及心理声学中的几大效应的介绍。并对现有的各心理声学参量数学模型进行了详细的介绍，包括稳态的响度模型、两种常用的粗糙度模型、尖锐度模型以及波动度模型。分析了各模型的优缺点、数学原理和编程流程等。在了解学习了这些参量模型的基础上，利用matlab对这些模型进行了编程实现，并将所得的结果与LMS系统所得结果进行对比，发现这些模型可以有效地用于噪声声品质的评估。

然后，完成了进气噪声的主观评价实验。对主观评价实验的方法、主体以及实验条件进行了介绍，并详细地描述了实验的流程以及最终将所得主观评价结果进行验证。本文采用了国际上常用的验证方法对实验结果进行了验证，发现实验结果符合精度要求，准确性较高。

最后，本文就BP神经网络方面的知识进行了详细的介绍。包括BP神经网络的基本结构以及BP神经网络的学习算法和训练算法等。建立了基于BP神经网络的汽车进气噪声烦躁度预测模型，并利用上述工作中所得的数据进行检验，与之前计算所得的客观心理声学参量结果进行对比，发现本文所建的模型与客观心理声学参量有很好的联系，因此实现了噪声的主观评价结果的客观量化。

关键词：心理声学，声品质，神经网络，烦躁度模型，主客观评价

Abstract

Since the reform and opening up, the automobile industry has developed rapidly as a pillar industry of China's economy. At present, China is already one of the world's largest automobile producers. The development of the automobile industry and the improvement of living standards have led people not only to care about the practicality of the car, but also to pay more and more attention to car noise that affects the comfort of the car. After reading a large number of domestic and foreign literatures in the field of psychoacoustic research, it is found that the literature on the subjective and objective evaluation of automotive non-stationary noise is very scarce in the academic community. This is very necessary for improving the noise pollution of automobiles.

This article first analyzes the structure of the intake system of the car and the principle of the intake noise, and fully understands the relevant knowledge of the car's intake noise. Then, the experiment of measuring the intake noise of the car was carried out, and the original intake noise data signal was obtained. I used the self-made software of matlab and the post-processing software of LMS system to screen and post-process the noise signals obtained from the experiment. Finally, the processed sound files were internationally standardized, and finally the sound sample library used in this research was established.

Second, this article gives a detailed introduction to psychoacoustic knowledge, including several major effects in the human auditory system and psychoacoustics. The existing mathematical models of psychoacoustic parameters are introduced in detail, including steady-state loudness models, two commonly used roughness models, sharpness models, and volatility models. Among them, the volatility model refers to the model proposed by Zhang Wei, the president of Wuhan University of Technology, and introduces in detail the advantages of the new model, the mathematical principles, and the programming process. After introducing several commonly used models, I used matlab to program these models, and compared the results obtained with the results obtained by the LMS system and found that these models can be effectively used for the evaluation of noise sound quality.

After completing the above work, this article completed the subjective evaluation experiment of intake noise. First of all, this article describes the method, subject and experimental conditions of the subjective evaluation experiment, and then describes in detail the process of the experiment and the verification of the final subjective evaluation results. In this paper, we use the commonly used verification methods in the world to verify the experimental results and find that the experimental results meet the requirements and the accuracy is high.

Finally, this paper gives a detailed introduction to the knowledge of neural network. It includes several basic structures of neural networks and learning algorithms for neural networks. After learning the relevant knowledge of neural networks, this paper establishes a predictive model of car intake noise irritability based on BP neural network, and uses the data obtained in the above work to test, and with the previously calculated objective psychoacoustic parameters results In contrast, it was found that the model built in this paper has a good relationship with objective psychoacoustic parameters. Therefore, objective quantification of subjective evaluation of noise is achieved.

Keywords: Psychoacoustics, Sound quality, BP neural network, Annoyance model, Subjective and objective evaluation

目录

摘要 I

Abstract II

第1章 绪论 1

1.1课题的研究背景及意义 1

1.2 国内外研究现状 2

1.2.1 国内声品质研究现状 2

1.2.2 国外声品质研究现状 3

1.3 声品质研究成果的应用现状 4

1.4论文主要研究内容 4

第2章 汽车进气噪声实验 6

2.1 汽车进气噪声的产生原理及进气系统结构 6

2.2 汽车进气噪声的测量 6

2.2.1 实验所需设备 6

2.2.2 实验条件 7

2.2.3 噪声数据采集方案 7

2.2.4 样本数据处理 8

2.3 本章小结 8

第3章 心理声学原理及其参量算法 9

3.1 人体听觉概述 9

3.1.1 听觉系统的构造 9

3.1.2 听觉掩蔽效应 10

3.1.3 双耳效应 11

3.1.4 临界频带 11

3.2 A计权声压级 13

3.3 Zwicker 稳态响度模型 14

3.4 尖锐度模型 16

3.5 粗糙度模型 17

3.5.1 DW 与 Aures 粗糙度模型的区别 17

3.5.2 DW粗糙度模型 18

3.6 波动度模型 21

3.7 模型的验证分析 23

3.8 本章小结 24

第4章 进气噪声主观评价试验 25

4.1 进气噪声主观评价试验 25

4.1.1 主观评价方法 25

4.1.2 主观评价主体 27

4.1.3 声音样本 28

4.1.4 听音环境 28

4.1.5 评价结果的数据检验 29

4.1.6 主客观评价结果 30

4.2 本章小结 31

第5章 进气声品质预测模型的构建 32

5.1 BP 神经网络 32

5.1.1 BP 神经网络结构 32

5.1.2 BP 学习算法 33

5.2 烦躁度 BP 网络预测模型 34

5.2.1 输入输出数据归一化处理 35

5.2.2 输入输出层节点数的确定 35