SVM-based sound classification based on MPEG-7 audio LLDs and related enhanced features

Abstract

In this paper, we present a support vector machine (SVM) based sound classifier using MPEG-7 audio low-level descriptors and related enhanced features. The SVM is used to construct an environmental sound classifier. As for the audio features, a feature set which includes MPEG-7 audio low-level descriptors, spectrum centroid, spectrum spread, and spectrum flatness, and two enhanced features, RSS and Legendre-based trend coefficients (LBTCs) of the spectrum power is adopted for sound classification. Experiments demonstrate the proposed system can achieve an 81.25% classification rate. © 2012 Springer-Verlag.