SOUND SOURCE CLASSIFICATION USING SUPPORT VECTOR MACHINE

This paper shows an application of a learning method for acoustic signal classification by an auditory robot. The learning approach provides an unified acoustic signal classification method without considering the characteristics of target signals. Support Vector Machine was adopted to obtain the classifier and the target signal was characterized by Mel-Scale Log Spectrum which was a general form to symbolize acoustic signals. Results of actual experiments to classify 4 class of acoustic signals at single sound source case and to classify 3 class of acoustic signals at plural sound source case showed the validity of the method.