Live Event Detection for People Safety Using NLP and Deep Learning

In today’s world, human actions such as robbery, assault, and homicide pose significant threats to society, especially to individuals working alone at night in remote areas—women being particularly vulnerable. These threatening incidents are often accompanied by distinct sounds or noises, which can serve as early indicators of danger. Although several existing security measures are available, many fall short due to their limited accuracy and delayed threat detection. To address this, a novel software-based prototype has been developed that detects potential threats in real-time by analyzing ambient sounds or noises. Without the need for any additional hardware components, the system can automatically alert a victim’s pre-registered contacts by sending notifications via email, SMS, and WhatsApp directly through their smartphone. Audio signals sourced from a Kaggle dataset were visualized and examined using Exploratory Data Analysis (EDA) techniques. The insights obtained from EDA were then used to train deep learning models, including Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), achieving an impressive 96.6% accuracy in classifying various audio events.