Audio Distress Signal Recognition in Rural and Urban Areas using a WSN consisting of Portable Resource-Constrained Devices

The safety of women, children and old-aged individuals in rural and urban areas is at great risk in many countries. With an unprecedented rise in population of cities & villages and the huge adoption of technology, it is vital to create new efforts toward safeguarding urban and rural residents against crime. Improving emergency systems in rural regions & big cities has been one of the biggest hurdles faced today and very little work has been done in this area. To address safety concerns, we developed a technology framework to detect and report distress signals with very low response times. Speech processing and machine learning techniques were employed to classify sounds in real-time, with varying levels of background noise. To ensure scalability and low-cost deployments of the device in rural/urban areas, the machine learning model has been modified to run inferencing tasks on power and memory constrained embedded devices. A WSN has also been proposed to span large regions with very short response times. The developed framework gives us high accuracies, in the order of 98% and small response times, in the order of a few milliseconds. This allows the device to focus more on routing the data to a WSN cluster head in a shorter amount of time, which could then be reported to the base station operated by the governing security authorities.