Indoor Sound Classification with Support Vector Machines: State of the Art and Experimentation

Abstract

Sound classification is considered as one of the most important areas of classification domain, but the least developed compared to speech and voice recognition. In this study, we focus on the works that deal with sound classification by making a comparative study based on feature extraction and classification methods as well as the targeted sound corpus. Next, we present an overview of sound classification systems that utilize deep learning techniques, aiming to compare them with traditional learning methods. Based on our previous studies and conclusions, and considering that the challenge in choosing classification methods lies in balancing accuracy and computational cost, we conducted experiments using SVMs (support vector machines) with different kernels and MFCCs (Mel frequency Cepstral coefficients). Tests are carried out for the classification of some indoor abnormal sounds, then the number of classes is increased to cover a wider variety of sounds in order to observe and study the system's behavior. Finally, the results obtained in this work are promising and motivate us to explore deeper tests which are mentioned in the discussion and conclusion section.