Environmental sound classification using spectral dynamic features

Abstract

This paper presents a novel feature extraction method for environmental sounds classification. Although many kinds of audio features have been investigated for environmental sound classification tasks, most of them have been extracted only to model the speech signal structure, which explains their lower performance when dealing with other kinds of audio signals. The method proposed in this paper processes and extracts the spectral changes throughout the frames of a sound file and appends them to the frame-based spectral feature vectors as dynamic features. Experimental results show that proposed features outperform the commonly used audio features in context recognition tasks.