Urban Sound Classification using VGG19 Convolutional Neural Network (CNN) Model and its Visualisation

This research aims to categorize urban noise effectively through the application of a VGG19 Convolutional Neural Network (CNN) model, a robust deep learning framework designed for processing audio signals. Accurate identification of urban sound is crucial for public safety, environmental monitoring, and the progression of smart cities. The proposed method utilizes the VGG19 CNN architecture to rapidly extract hierarchical components from audio inputs, ensuring accurate recognition of a diverse array of urban noises. Our study involves accumulating a comprehensive collection of urban sounds, encompassing various ambient noises commonly encountered in cities, which is employed to train the VGG19 CNN model. A notable strength of this model lies in its autonomous ability to learn hierarchical representations of audio data. Through rigorous testing based on predefined criteria, we evaluate the model's capacity to distinguish between different urban sound classes. To gain insights into the internal representations acquired by the CNN model, the study employs visualization techniques, such as heatmaps generated using gradient-weighted class activation mapping (Grad-CAM). The results showcase the practical efficacy of the VGG19 CNN model, achieving an 86% accuracy rate in classifying urban noises. The integration of visualization tools enhances trust and acceptance in real-world applications by elucidating the model's decision-making process. This research contributes to the advancement of intelligent systems capable of assessing city sounds, emphasizing the importance of combining interpretability tools with deep learning models for increased reliability and user-friendliness.