Using single log-Gabor filter and reassignment method for audio classification applications

We present a robust environmental sound classification approach, based on reassigned spectrogram and log-Gabor filters. In this method, reassigned spectrograms are passed through an appropriate log-Gabor filter and the outputs are underwent an optimal feature selection procedure based on mutual information criteria. The evaluation of this classification system is performed on a corpus of 6 environmental sounds classes. The best performance was obtained using multi-class support vector machines (SVM's), producing an average classification accuracy of the order 91.77%.