Research on speech distortion measure based on neural network

This paper puts forward a novel spectrum distortion measure (MSD) of speech based on Artificial Neural Network (ANN). It provide the MSD with the capability of employing subjective behavior of perception of human's auditory system, by using the theory of approaching to nonlinear function of multilayer perceptrons (MLP) and radial basis function networks (RBFN). In the application of speech quality objective assessment, the result of subjective evaluation under plenty of distortion conditions is directly used as the desired value of training. Statistic analysis shows that, comparing with traditional Euclidian distance and weight Euclidian distance, the correlation of subjective to objective of MSD based on ANN has great improvement, also are the reliability and robustness. Moreover, it has the property of technology independent.