Identification of machinery sounds

Abstract

Speaker recognition methods are well known and widely used in the ASR (Automatic Speech Recognition) systems. The use of these methods for the classification of machinery sounds in noisy environments is presented in this paper. Influence of background noise was reduced by using a highly directive sound recording, which can be understood as a spatial filter. A fusion of microphone antenna with beamforming algorithm forms such a filter, which improves SNR (Signal to Noise Ratio). Features of machinery sounds have been extracted using standard MFCC (Mel Frequency Cepstral Coefficients) parameterization method with Mel and linear frequency scaling. SVM (Support Vector Machine) classifier was used for the classification of sound features. A significant improvement of the classifier decision performance was achieved in noisy environment when 8 microphones were used together with beamforming algorithm. Results of using Mel and linear scale are also presented and show similar results in recognition of machinery sounds.