International Research Journal of Education and Technology

The increasing crime rates and delayed responses to emergencies highlight the need for advanced automated systems capable of detecting distress signals in real-time. This project addresses the issue by developing a sound-based detection system that utilizes machine learning and deep learning to recognize human screams in public spaces, especially during emergencies. By leveraging audio classification techniques, such as Support Vector Machines (SVM) and Multilayer Perceptrons (MLP), the system can distinguish screams from other sounds, minimizing false positives and ensuring high accuracy in real-world conditions. The proposed system operates continuously in the background on devices equipped with microphones. When a scream is detected, the application evaluates the level of risk. If the sound meets the criteria for a high-risk situation, the system sends an automated alert message, including GPS location, to the nearest authorities via SMS, enabling quick response. This project’s objective is to improve public safety by reducing the time required for authorities to respond to potentially life- threatening situations, ultimately contributing to a safer society.

The following report outlines the objectives, methodology, system architecture, and implementation of the scream detection system, with an emphasis on accuracy, real-time performance, and user privacy. This application has the potential to transform public safety measures and pave the way for broader AI-based security solutions  in urban environments.

Key Words: Scream detection, Machine learning, Deep learning, Audio classification, Public safety, Support Vector Machines, Multilayer Perceptrons, Real-time alert system.