Development of a real-time panic alert system using threshold-based audio detection

This study presents the development and evaluation of an IoT-based real-time panic alert system for emergency situations using a threshold-based audio detection method and mobile network communication. The system integrates a voice sensor, Arduino micro-controller, GSM module, cloud-based real-time database, and an Android mobile application to enable automatic panic detection without requiring manual user activation. Short audio frames are continuously captured, and their amplitude values are compared against a predefined panic threshold derived from literature-reported panic-level vocal intensities and experimental testing. System performance was evaluated under the conditions: quiet room, office noise, and street noise, using both human voice input and a calibrated 100 dB sound source. Performance testing was conducted over 2G, 3G, and 4G mobile networks across 100 trials per network type. The results show mean alert latencies of 3.8s, 2.1s, and 1.2s for 2G, 3G, and 4G networks, respectively, with corresponding reliability values of 85%, 92%, and 97% for alerts delivered within 10s. The system recorded a false positive rate of 7% and a false negative rate of 9% during testing. The results demonstrate that combining a simple threshold-based audio detection approach with IoT connectivity and GSM messaging can provide timely and reliable panic alerts, particularly in scenarios where users may be unable to manually trigger emergency notifications.