MFCC and CNN-Based Sound Classification Method for Detection of Forensic Cases in Smart Cities

Abstract

At present, security surveillance systems relying primarily on visuals may have gaps relating to actual security breaches. Hence, active responses to a variety of security risks are provided in an unprecedented manner by utilizing an anomaly detection system based on sound. Augmented sound input to surveillance cameras will be of utmost importance for providing timely assistance in cases of emergencies. In this project, the sound classes include notable or anomalous sound events like gunshots, glass shattering, brake sounds, and human conversation in a bustle that are integrated to work in tandem with security cameras. By recognizing not only sounds that represent security risks but also normal noise of a crowd, the system is expected to use deep learning and sound recognition methods to effectively reduce the false alarm rates and improve accuracy. This method encourages the creation of multifunctional camera devices that are cost and time effective. © 2025 IEEE.