Characterization of Internet of Things (IoT) powered-Acoustics Sensor for Indoor Surveillance Sound Classification

Abstract

In this research, the focus area is on the surveillance sound detection from the Internet of Things (IoT) system level perspective, which is to investigate the accuracy of the sound classification by using deep learning methods concerning the sound level. This is translated to the sound event classification concerning the distance of the sound source. In this research, the AclNet model is used for sound event classification and MEMs microphone manufactured by Knowles (SPH1668LM4H-1) is used to record the sound event. The actual sound pressure level of surveillance sound has been considered. From the research observation, by using a single microphone arrays, and the sound classification confidence level’s threshold set at 0.4, AclNet model can classify accurately up to the distance of 2.5m, 6.3m, and 158.5m for baby crying, siren, and gunshot sound events respectively. These results provide an insight into the area of coverage for surveillance sound classification by using IoT acoustics sensor.