Detection And Identification Of Special Point Source Sound Signal

This relatively stable society still exists some hazards such as gun crime and terrorists. It is a reasonable way to locate sound source bases on sound wave propagation law. Its essence can be attributed in mathematical inverse problem. This subject is corpus of a research project commissioned by a communication technology company in Nanjing. It realizes the detection and recognition of gunfire or explosion danger signal based on the analysis method of research the acoustic parameters such as energy, frequency. It establishes the necessary theoretical basis for the study of the sound source localization system.Usually shooting process will produce two forms of sound wave, the first is Muzzle blast produced by the gunpowder exploded, the second is shock wave produced when the bullets flying in the air with supersonic speed. Sound localization system fixes the position of gunfire bases on the two waves. The main content of this study is to test and identify the two waves.First, it shows the basic principle of pretreatment, detection and recognition of the two waves, and then the detection and identification process of Muzzle blast and Shockwave signals was studied, identified detection and identification method for the two waves, and gives the result analysis. For the Shockwave signal is a sound signal, its energy and frequency is high, so it can be detected and identified directly. Muzzle blast signal’s frequency and energy is lower, it is vulnerable to the interference of noise and Shockwave reflection. Use the wavelet denoising method to filter Muzzle blast before its detection in order to improve the detection efficiency. The two waves are detected based on signal to noise ratio. It gives the calculation method of average noise power and the detection threshold. Finally, it adopts the method of GMM to identify the voice signal, using the characteristic parameters of short time average energy, MFCC and its differential coefficient. By choosing different characteristic parameters to train the model and analyze the recognition results, it gets the effect of characteristic parameters on the result of recognition. Then by selecting different wave signal to detect and identify gunfire, gets the effect of wave selection on the result of recognition. Analysis results show that the recognition rate of algorithm in this paper is high, it is an ideal recognition method.