Research On The Recognition Method Of Impulse Acoustic Wave Signal

Acoustic signals contains many useful information,especially the impulsive signals which have short duration but bursts of short-time energy,it often indicates risk events,such as gunshots,explosions,broken glass,and human screams.The recognition of impulsive signals can warn the critical situation and help to safeguard people's personal and property safety.As the typical representative of the impulsive signals,the detection of the gunshot can help maintain the stable social security.The gunshot signals include Muzzle Blast and Shock Wave,and this paper mainly studies the recognition method of Muzzle Blast as the representative for impulsive signals.First compare the time and frequency characteristics of impulsive signals and intermittent signals,choose two features that is commonly used in speech signal recognition:the wavelet energy and MEL cepstrum coefficient(MFCC),then extract the feature and use support vector machine(SVM)for recognition.The experimental results show that the intermittent signal and impulsive signal vary greatly,recognition methods cannot be directly transplant.Then use MFCC feature applying to impulsive signals to identify Muzzle Blast.MFCC can not well distinguish the Muzzle Blast and non-Muzzle Blast signals,so we put forward a special characteristic of short-time Energy and short-time Energy Ratio to recognize Muzzle Blast,recognition rate reached over 98%,and the discrimination between Muzzle Blast and Shock Wave,voice sound is pretty high.Above recognition is aimed at the Muzzle Blast recognition of the sample fragment,and then,the detection of the Muzzle Blast of continuous monitoring data based on the recognition model of the Muzzle Blast is carried out.First,extract features respectively based on three data processing methods of same scale frame of before and after positioning starting point of Muzzle Blast and mutative scale frame after positioning starting point of Muzzle Blast,then take the Muzzle Blast detection of continuous monitoring data.The experimental results show that the first method has many false alarm and the detection rate is 65.72%,the second method has false alarm reduction and the detection rate is 71.43%,the third method has no false alarm and the detection rate is 100%.It proved that Muzzle Blast is sensitive to its starting point position and the processing method of variable scale frame retains the fast time-varying characteristics of the Muzzle Blast.