Gunshot detection in noisy environments

Gunshot detection finds application in the fields of law enforcement, forensic science, and defense (military applications). The first task of a sniper detector, aiming to estimate the direction of arrival of a given gunshot, is to detect automatically the presence of this audio event. Since it is a typical on-line application where a fast response is of paramount importance, a non (computationally) expensive procedure is needed. In a recent work, a simple procedure based on the correlation of the audio signal against a template has proved its efficiency as a gunshot detection algorithm. In this paper, we extend its evaluation to a noisy environment and assess its performance, in gunshot recognition and gunshot detection tasks, comparing it to other more complex methods. Keywords— gunshot detection; gunshot classification; impulsive signals; Hidden Markov Models; Mel-frequency cepstral coefficients; stable distributions; linear predictive coding.