The Improvement Performance of Speaker Verification System Through the Multi-Vector Quantization Codebook Structure

멀티 VQ 코드북을 이용한 화자확인 시스템의 성능개선

In this paper, we propose the new method that separate the existing common VQ code book into two parts, one is the common VQ code book which is the half of existing common VQ code book, another is the personal speaker VQ code book which accommodate the personal speaker characteristic, variation to improve the performance of the text-dependent speaker verification system using discrete HMM. We apply the propose method m this paper to the text-dependent speaker verification system using discrete HMM and have the improvement performance of about 0.24% compared to existing method