Distress Detection Using a Hybrid SVM - CNN Classifier

Due to the escalating frequency of reported crimes, there has been a surge in research endeavors focused on various approaches to enhance monitoring and surveillance techniques. As opposed to vision-based applications, which are currently the most used framework for monitoring purposes, audio-based systems can be more flexible and relatively less intrusive. While current research studies predominantly utilize images as the primary input for Deep Learning (DL) algorithms, it is worth noting that sound can also serve as a valuable source of input for these models. In this paper, we propose and develop a novel hybrid deep learning model for identifying and detecting people in distress by their screams. The working of our proposed system is built by integrating sound detection module and DL model which will help us to detect if a person is in distress or not. The system uses hybridization concept consisting of Support Vector Machine (SVM) and Convolutional Neural Network (CNN) models, with the audio snippet undergoing 3 levels of classification, with the accuracy of each level found to be 93%, 100% and 92% respectively. The audio data is finally classified as an instance of distress or no-distress.