Abstract: Audio is an important medium that carries substantial information to meet human needs. To improve accuracy of audio classification, we propose a new algorithm with Mel frequency cepstrum coefficient (MFCC) parameters as the feature vectors, and use a back propagation (BP) neural network model based on improved transfer function to classify six types of audio signals. Experiments show that the proposed algorithm has good performance and the improved transfer function converges faster that the traditional BP algorithm. It can reduce training time, and improve classification accuracy up to more than 90%.

Key words: audio classification, BP neural network, classification accuracy, convergence speed, transfer function