Classification of human sounds using support vector machine with psychoacoustic data

Abstract

This paper presents the classification of human sounds based on support vector machine (SVM) using psychoacoustic data. A scream classification model, with sounds of speech and screams indicating different acoustical characteristics, was investigated. Temporal changes were observed by evaluating the physical characteristics of waveforms and spectrograms with psychoacoustic parameters, including loudness and sharpness. Mel frequency cepstral coefficients were used to identify the spectral energy distribution of screams. Further, a Mel filter bank and frequency band filter were used to extract the high spectral energy, and differentiate between the lower and higher energy spectra. The classification accuracy was improved by combining the SVM with the psy-choacoustic parameters of scream sound.