Speaker Identification Based on Weighted FSVQ

This paper proposes a method of weighted finite state vector quantization (WFSVQ) to address the issue existed in traditional vector quantization.The issue is that the recognition rate of traditional vector quantization is low when it has small codeword number.A weighted FSVQ combines the static characteristic of speech with time correlation (dynamic characteristic) of speech frames.It calculates quantization distortion twice and weights them according to their contribution and quantitative accuracy.The weighted sum is used as the final judgment.The experimental results show that this method is superior to traditional vector quantization.Especially in the small codeword number (not more than 8),the recognition rate increases more than 10%.