基于MFCC与改进ACF的汽车声音识别算法研究

摘要

Vehicle audio recognition is the foundation of vehicle sound source localization and other automotive research,it is very impor-tant for traffic accidents identification,crime scene evidence and crime reduction. The problem of high computational complexity and rela-tively low recognition rate has existed in current vehicle audio recognition. Concerning those problems above,the vehicle recognition algo-rithm taking Mel-Frequency Cepstrum Coefficients and improved Auto-Correlation Function as hybrid feature is applied in the vehicle audio recognition system. Modeling and classifying by the Gaussian Mixture Model,this feature vector outperforms MFCC and Differenti-al MFCC features in recognition. The simulation results prove the effectiveness of the proposed algorithm.%汽车声音识别是汽车声源定位等研究的基础，对交通事故鉴定、犯罪举证和犯罪现场还原等具有重要意义。现有汽车声音识别算法存在算法复杂度高和识别率相对较低等问题。针对现行问题，将以梅尔倒谱系数( MFCC)特征与自相关函数(ACF)方差作为混合特征的汽车声音识别算法应用到汽车声音识别系统中。该算法使用高斯混合模型(GMM)进行汽车声音建模和识别，获得比MFCC特征及其一阶差分特征组成的混合特征更好的识别效果。并通过仿真实验证明了该算法的有效性。