基于特征参数LPCC与AMDF的异常声音检测  

Abnormal Audio Detection Based on Feature Parameters LPCC and AMDF

作　　者：许文杰 杨淇善 XU Wenjie;YANG Qishan(Faculty of Infonnation Technology,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124

出　　处：《长江信息通信》2021年第10期110-113,共4页Changjiang Information & Communications

摘　　要：区别于视频监控这一重要技术手段,异常声音从监控者的听觉角度十分高效的反映出监控场景的异常情况,对于监控公共场所的安全有着十分显著的作用。因此,异常声音的研究方法也受到了国内外许多学者和机构的重视。文章提出了一种由线性预测倒谱系数(Linear Prediction Cepstrum Coefficient,LPCC)与短时平均幅度差(Average Magnitude Difference Function,AMDF)融合的新型联合特征参数并结合高斯混合模型(Gaussian Mixture Model,GMM)应用于异常声音检测工作,实验结果显示所实现方法在对危险声音检测时得到了较好检测准确率。differing from video surveillance,which is an important technical tool,abnormal audio is very efficient in reflecting anomalies in surveillance scenes from the perspective of the monitor’s hearing,and has a very significant role in monitoring the safety of public places.Therefore,the research method of abnormal audio has also received the attention of many scholars and institutions at home and abroad.In this paper,we propose a novel feature parameter consisting of Linear Prediction Cepstrum Coefficient(LPCC)fused with Average Magnitude Difference Function(AMDF)and combined with Gaussian Mixture Model(GMM).GMM is applied to the detection of anomalous sound,and the experimental results show that the implemented method achieves better detection accuracy in the detection of dangerous sound.

关 键 词：异常声音 线性预测倒谱系数 短时平均幅度差 特征融合 高斯混合模型 

分 类 号：TN912[电子电信—通信与信息系统]