Special Sound Detection for emergency phones

Most emergency phones can be activated by a button press. However, to press the button requires the person needing help to touch the phone physically, which causes infeasibilities in practice: 1) it is hard for someone unfamiliar with the environment to locate an emergency phone quickly, and 2) even if the person knows where the phone is, she/he has to run a certain distance to reach it. Therefore, if an emergency phone is equipped with a sensor which can monitor special events such as a scream for help or a gun shooting, its capability of security surveillance will be increased tremendously. In this paper, we present a Special Sound Detection (SSD) algorithm we have successfully built and prototyped. The prototype is robust and yet simple to implement. The accuracy of detecting screams within a distance of 100 feet is 98.6%. For gunshots within 500 meters, the accuracy is 99.1%. The design of the algorithm is highly modular. It can be extended to detect other types of special sounds by adding relevant training samples.