Urban Sound Classification with Convolutional Neural Network

Tremendous population and urban growth rates all over the present potent threats to create livable and sustainable cities. It leads to the extension and transformation of urban noises, and the diversification of the latter is attributed to this increase. These sounds they turned into is still regarded more as noise than as actual information which is an important factor in the smart cities idea. This research seeks to establish whether it is possible to enhance the classification of urban sounds by adopting CNNs for sound recognition to help improve the safety of pedestrians by identifying possible danger sounds. According to the data, traffic injuries of pedestrians are escalating around the world, partly because of a lack of sensitivity to the surroundings. This work intends to categorize urban sounds, for instance sirens or noise produced by vehicles in real time so as to enlighten the pedestrian of close dangers, hence avoiding mishaps. The deep learning model built in this context is made up of three convolutional layers having a max pooling layer to improve the model's feature extracting ability and final classification accuracy with the GlobalAveragePooling2D layer. In this context, the performance of the model was assessed with UrbanSound8k, which consists of more than 8,700 urban audio samples, thus allowing for reliable performance analysis. Findings show that the model is quite effective at separating sounds commonly encountered in urban settings and reaffirms an intended use of the model which is to aid pedestrian safety. This research focuses on utilizing CNN-based audio classification in real-time hazard detection with possible extension to enhance urban safety and additional usage of sound-based alarms.