Contribution
Sound of Sirens as Acoustic Landmarks within the European Union
* Presenting author
Abstract:
This paper explores the use of deep learning models for classifying emergency siren sounds from four EU countries: France, Germany, Italy, and Spain. It introduces a VGG-based model for classifying sirens into categories such as police, ambulance, and fire truck, and compares its performance with a previously proposed ResNet model. The study addresses challenges like dataset imbalance and spectral variability through data augmentation techniques, improving model robustness. The paper highlights the importance of accurately classifying sirens for urban safety and emergency response. The VGG-based SICaPKF model demonstrates superior classification accuracy over the ResNet model, as indicated by confusion matrices and accuracy metrics. The study's findings emphasize the model's potential in enhancing emergency vehicle identification and urban acoustic monitoring, contributing to safer urban environments. The results suggest that deep learning models, particularly VGG architectures, can significantly advance siren detection and general sound classification in diverse urban settings.