Aportable system for robust acoustic detection of atypical situations

This work presents a practical, automatic and robust methodology for acoustic surveillance of hazardous situations. The proposed system efficiently identifies atypical situations which include scream, explosion and gunshot sound events under different kind of environments (e.g. metro station, urban etc). The main objective is to detect abnormal events which lead to life-threatening circumstances or property damage by helping an authorized officer to take the appropriate actions through a decision support interface. After extensive experimentations, a fully prob abilistic structure of Gaussian mixtures was designed which incorporates task depended feature sets. A testing procedure under different SNR conditions was followed and we report high detection rates with respect to false alarm and miss probability rates.