Emergency Vehicle Sound Classification for Dynamic Traffic Signal Control using Convolutional Neural Networks

Urban traffic management systems must ensure quick and efficient passage for emergency vehicles like fire trucks, ambulances and police cars to reduce response time during critical situations. This paper proposes an approach for accurately classifying emergency vehicle sirens using Convolutional Neural Networks (CNNs). The system utilizes a dataset of WAV audio files representing the sounds of ambulances, fire trucks, and police sirens. Mel-Frequency Cepstral Coefficients (MFCCs) were extracted as key features from the audio files to capture the essential characteristics of the sirens. Preprocessing steps included noise reduction and feature scaling to ensure the model's accuracy. A CNN was then trained to classify the audio into the three categories, achieving over 90% accuracy on the test set. This high level of accuracy demonstrates the system's potential for real-world application in intelligent traffic management systems, where the model can dynamically adjust traffic signals, providing priority access to emergency vehicles in real-time. Future extensions of this work will include integrating the system with live traffic data to improve the overall efficiency of urban mobility.