The Normalization Training Technique of State-Relative Direct Mean Shift Based on MAP Estimation

Speech normalization method is a technology that converts the spoken voice to machine-readable input. In this paper, we proposed a speaker normalization training technique based on model of mathematics statistics. This technique combined the normalization training technique of state relative direct mean shift with the method of MAP/WAR model adaptation into a robustness framework in order to provide a better original model for model adaptation technique, and also kept a balance between the increasing adaptation speed and keeping enough model smoothness. Finally, the experimental examination demonstrated that the method could improve robustness of speaker recognition in terms of supervised model.