A Noise Classification Method Based on Noise Energy Distribution in Bark Domain

By classifying various noises, the performance of the speech signal processing in the noise environments can be improved. In order to accurately distinguish various noises, this paper proposes a noise classification method based on the distribution characteristics of the noise energy in the Bark domain. By mapping the noise energy from the uniform time frequency space to the Bark space, this algorithm constructs a 22 dimensional feature vector that can effectively distinguish various noises. Moreover, the support vector machine (SVM) is utilized to train the model and classify the noises. Experimental results show that the proposed method has very high classification accuracy, and the average accuracies for two noise databases used in the experiments attain 99.50% and 93.44%.