SONICSENTINEL: AI-POWERED GUNFIRE EVENT RECOGNITION FROM AMBIENT SOUNDSCAPE

Abstract

Gunshot sound recognition has emerged as an important research area due to the increasing need for intelligent surveillance, public safety monitoring, and automated threat detection systems. Traditional gunshot detection methods often rely on manual monitoring or basic acoustic sensing techniques, which are limited in their ability to accurately distinguish firearm sounds from other environmental noises. Variations in recording conditions, background interference, and similarities between different weapon sounds further increase the complexity of reliable firearm identification. These challenges necessitate the development of an automated and intelligent system capable of accurately classifying gunshot sounds in real-world environments. To address this problem, the proposed system introduces a Machine Learning (ML)-based gunshot sound classification framework that analyzes firearm audio recordings and categorizes them into predefined weapon classes. The system performs audio preprocessing and extracts discriminative acoustic features such as Mel-Frequency Cepstral Coefficients (MFCC), chroma features, spectral contrast, and zero-crossing rate using the Librosa library. The extracted features are then utilized for classification through multiple ML algorithms, including Gaussian Naïve Bayes Classifier (GNBC), Logistic Regression (LR), Linear Discriminant Analysis Classifier (LDAC), and Extra Trees Classifier (ETC). Performance evaluation is conducted using accuracy, precision, recall, F1-score, classification reports, and confusion matrices to identify the most effective classification approach. A user-friendly graphical interface is developed using the Tkinter framework to facilitate dataset management, feature extraction, model training, evaluation, and firearm prediction. The system supports role-based authentication for administrators and users, enabling secure access to various functionalities.