Speaker Identification Based on VQ and Square Deviation Amending

摘要

A method for Text-Independent speaker Identification is proposed in this paper. The proposed method is based on VQ(Vector Quantification) and modified by variance. This paper analyzes theoretically the feasibility and deficiency of the VQ based model and proposes an algorithm to remedy these defects by variance. Meanwhile, this paper also shows how to select feature parameter through both theoretical analysis and experiments. In addition, the concept of average mutuality-self variance projection is introduced as well as a formula to quantify the ability of identification of feature vector's every dimension. Experiment results show that the proposed speaker identification method achieves very high identification rate.