基于小波包分析的玻璃破碎声音识别系统设计  

Design of Glass Breaking Sound Recognition System Based on Wavelet Packet Transform

作　　者：李颀 白雨尼 王丹聪 

机构地区：[1]电气与信息工程学院

出　　处：《计算机测量与控制》2018年第1期168-172,共5页Computer Measurement &Control

基　　金：西安市未央区纵向科技计划项目(CXY1129);陕西科技大学博士科研启动基金(BJ13-15)

摘　　要：针对传统家居安防监控系统中多采用声音振动传感器对玻璃破碎进行检测,不仅安装不便且不美观,因此提出了一种非接触式玻璃破碎声音识别系统;目前常用FFT(傅里叶变换)对声音信号进行分析处理的缺点是不能显示出声音频率出现的时刻,因此,提出一种基于小波包分析的玻璃破碎声音识别系统;通过对玻璃破碎声音信号采集、预处理;随后利用小波包分析代替FFT方法,提取小波包系数特征参数,并融合短时平均幅度和短时过零率,从时域和频域两个角度的特征参数来表征玻璃破碎声音;最后利用隐马尔科夫分类器(HMM)从提取到的特征参数中训练出玻璃破碎声音的HMM模型,对玻璃破碎声音进行识别;通过实验验证,该系统的识别率可达93%。Sound vibration sensor is often used to test the broken glass in traditional home security monitoring system,and this method not only is inconvenient installation but also is not beautiful.So this paper puts forward a kind of non-contact broken glass voice recognition system.The disadvantage of the commonly used FFT(Fourier transform)which is used to analyze and process the voice signal is not able to show the moment when sound and frequency appeared.Therefore,this paper proposes a glass breaking voice recognition system based on wavelet packet analysis.At first,this method should acquire and preprocess the signal of glass breaking voice,then using wavelet packet analysis instead of FFT method to extract wavelet packet coefficient characteristic parameters and fuse short-term average amplitude and short-time zero crossing rate,it also represent the glass breaking voice from two angles of characteristic parameters of time domain and frequency domain.Finally,using hidden Markov classifier to train hidden markov model(HMM)model of glass breaking voice from the extract characteristic parameters,then recognize glass breaking voice.Through experimental verification,this system recognition rate can reach 93%.

关 键 词：家居安防 特征提取 小波包分析 隐马尔科夫分类器 

分 类 号：TN3[电子电信—物理电子学]