基于卷积神经网络的汽车行人警示音评价系统设计  

Design of Automobile Pedestrian Warning Sound Evaluation System Based on Convolutional Neural Network

作　　者：于佳[1] 王得天[1] Yu Jia;Wang Detian

机构地区：[1]泛亚汽车技术中心有限公司,上海200135

出　　处：《时代汽车》2024年第13期122-124,共3页Auto Time

摘　　要：为了使电动汽车的行人警示音符合人耳主观感受及汽车品牌定位,设计了一个基于卷积神经网络的声品质评价系统,实现了对行人警示音频的客观评价。采用等级评分对设计好的音频文件进行主观评价,并获得主观评分。基于ISO 532-1:2014标准计算音频文件的响度、粗糙度、抖动度、烦扰度、尖锐度等声品质客观参数,并将其作为卷积神经网络模型的特征输入。评价模型的输出设定为豪华,舒适,科技三个指标。经过数据训练,模型可以有效输出给定指标的评价分数,并与主观评价分数吻合良好。所提出的模型可以实现端到端的声品质客观评价,评价结果能够有效反映人耳主观感受,从而为行人警示音的快速评价提供新的方法。To make the pedestrian warning sound of electric vehicles conform to the subjective perception of human ears and the brand positioning of automobiles,a sound quality evaluation system based on convolutional neural network is designed to realize the objective evaluation of pedestrian warning audio.Grading is used to subjectively evaluate the designed audio file and obtain a subjective score.Based on the ISO 532-1:2014 standard,the objective parameters of sound quality such as loudness,roughness,jitter,annoyance,and sharpness of the audio file are calculated,and the feature input of the convolutional neural network model is used.The output of the evaluation model is set to three indicators:luxury,comfort,and technology.After data training,the model can effectively output the evaluation score of a given indicator,and it is in good agreement with the subjective evaluation score.The proposed model can realize the end-to-end objective evaluation of sound quality,and the evaluation results can effectively reflect the subjective perception of the human ear,so as to provide a new method for the rapid evaluation of pedestrian warning sounds.

关 键 词：行人警示音 声品质评价系统 声品质参数 卷积神经网络 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]