In this project report we will present some basics about sound analysis that are relevant to our classification approach, the nature of our selected dataset and interesting findings from existing literature.
Environmental/Urban sound classification is a new subject and not much research is done comparing to music genre classification not to mention image classification. However there are some interesting papers that are dealing with urban sound analysis and classification that helped us navigate to this new upcoming topic. [3][4][5]
It seems that the most popular and performant solution of urban sound classification is to extract different kind of features from raw audio data into images and then perform image classification. Given this main idea some interesting proposed solutions exist in literature that try to experiment with combinations of different approaches.
From our research sound classification through image classification using convolutional neural networks seems to be the most efficient approach. Using MEL spectrograms, MFCC spectrograms, Chromagrams and Convolutional Neural Networks will be our main approach of implementing our urban sound classification.