基于MFCC的汽车敲击异响识别  

Recognition of vehicle’s rattle based on MFCCs

作　　者：黄凯[1] 郑瑶辰 邓兆祥[1,2,3] HUANG Kai;ZHENG Yaochen;DENG Zhaoxiang(College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400044,China;China Automotive Engineering Research Institute Co.,Ltd.,Chongqing 401122,China;State Key Lab of Vehicle NVH and Safety Technology,Chongqing 401122,China)

机构地区：[1]重庆大学机械与运载工程学院,重庆400044 [2]中国汽车工程研究院股份有限公司,重庆401122 [3]汽车振动噪声与安全技术国家重点实验室,重庆401122

出　　处：《振动与冲击》2022年第13期275-282,共8页Journal of Vibration and Shock

摘　　要：现阶段,汽车异响的诊断主要依赖有经验的工程师进行主观评判,存在不准确、易错判、易漏判的问题。针对汽车敲击异响实测信号进行统计分析得到梅尔倒谱系数(Mel frequency cepstrum coefficient,MFCC),并以此作为表征异响来源的特征向量,基于最大似然估计法构建其联合概率分布高斯混合模型(Gaussian mixture model,GMM),从而针对未知实测异响信号可利用该GMM模型进行似然判别。指出了说话人识别技术与敲击异响识别的不同之处即Mel三角滤波器个数和离散余弦变换输出系数个数的选取方式,并对方法的可行性进行分析,最后试验加以验证。结果显示此方法的识别率达100%,拒绝率达100%以上,为汽车异响的客观评价方法打下基础。At present,diagnosing vehicle squeak&rattle mainly depends on subjective judgment of experienced engineers with inaccurate,easy mis-judgment and easily missed judgments.Here,the measured signal of automobile knocking noise was statistically analyzed to obtain Mel frequency cepstrum coefficients(MFCCs),MFCCs were used as feature vector to characterize the source of abnormal noise.Based on the maximum likelihood estimation method,its joint probability distribution’s Gaussian mixture model(GMM)was constructed.Then,GMM model was used for likelihood discrimination of unknown measured abnormal sound signal.It was shown that the difference between speaker recognition technique and abnormal knocking sound recognition is the choosing method for number of Mel triangular filters and number of discrete cosine transform(DCT)output coefficients.the feasibility of the choosing method for the latter was analyzed,and finally it was verified using tests.The results showed that both recognition rate and rejection rate of this choosing method are 100%;this method lays a foundation for objective evaluation method of vehicle squeak&rattle.

关 键 词：说话人识别 敲击异响 梅尔倒谱系数(MFCC) 高斯混合模型(GMM) 

分 类 号：TP274[自动化与计算机技术—检测技术与自动化装置] U467.4[自动化与计算机技术—控制科学与工程]