Audio Feature Combination For Environmental Sound Classification

Abstract

Environmental Sound Classification (ESC)  is  one  of  the  most  trending research areas. ESC is classifying different  environmental  sounds for context-aware applications. ESC is considered complex in comparison to speech and music processing owing to the unstructured nature of environmental sounds. In the past, researchers have worked on ESC. Certain preprocessing techniques, features, and classifiers are explored. Though considerable accuracy is attained, the performance can be improved by combining different features. In this paper, a few features are combined and classification is performed using five machine learning classifiers. It is found that the combination of cepstral and spectral fea- tures (Mel Frequency Cepstral Coefficients and Chroma features) gives the best accuracy of 81.54% with the K-Nearest Neighbor classifier.