Live Event Detection for People’s Safety Using NLP and Deep Learning

Abstract

This project presents a real-time abnormal sound detection system for public safety using a hybrid Convolutional Neural Network–Long Short-Term Memory (CNNLSTM) architecture. The system continuously listens to environmental sounds and identifies potentially dangerous events such as screams, explosions, gunshots, or violent disturbances. Unlike traditional single-stage models, the proposed CNNLSTM approach combines the spatial feature extraction capability of CNNs with the temporal sequence learning strength of LSTMs, enabling effective recognition of complex and noisy sound patterns. Captured audio signals are converted into melspectrograms and fed into the CNN-LSTM model to extract deep acoustic features for highly accurate abnormal-event classification. When a threat is detected, the system instantly alerts users, ensuring quick response during critical situations. This method reduces false alarms, enhances robustness in real-world noisy environments, and achieves strong performance without requiring extremely large manually labeled datasets.