Analysis of Discriminative Vector Quantization Approach for Speaker Identification

摘要

Discriminative Vector Quantization method for Speaker Identification (DVQSI) considers the distribution of the interspeaker variation inside the speech feature vectors. When the parameters of DVQSI are suitably selected, the DVQSI technique yields better Speaker Identification (SI) accuracy than that of the existing Vector Quantization (VQ) technique. In this work, various techniques for speech feature vector space segmentation, discriminative weights assignment, and discriminative weighted average distortion pairs calculation, associated with DVQSI, are introduced. The performance of DVQSI by employing the proposed techniques is analyzed and tested experimentally. The experimental results confirm the SI accuracy improvement employing the proposed approach.