Analysis of linear predictive coefficients for gunshot detection based on neural networks

This work deals with analysis of the linear predictive coefficients with respect to their use for acoustic gunshot detection. First, coefficient stability was observed when changing length and position of an analysed signal frame. Then, the optimal prediction order was investigated. Finally, false alarms and correct gunshot detections were tested for various numbers of coefficients. The best experimental results achieved were 8.6% false alarm rate and 88% gunshot detection rate.