Sound Classification Algorithms for Indoor Human Activities

Abstract

The goal of this paper is to perform a comparison on different classification algorithms applied on Mel-Frequency Cepstral Coefficients and Moving Picture Experts Group-7 features in order to obtain a high average correct classification rate, greater than 98%, and a low computation time, less than 1 minute, for audio classification purposes in the case of audio signals from service robots. The highest correct classification rates are obtained using the Linear Discriminant Analysis for classification phase. For Mel-Frequency Cepstral Coefficients the averaged accuracy is 99.78%, using 64 features, and the classification computation time is 5.26 seconds. For Moving Picture Experts Group-7 features the averaged accuracy is 99,65%, with a classification computation time of 5.30 seconds.