多级矢量量化语音波形

MULTISTAGE VECTOR QUANTIZATION OF SPEECH WAVEFORM

本文介绍了一种通用多级矢量量化器的设计方法。这种量化器能有效地克服运算复杂度与存贮器需要量随矢量维数成指数律增长的困难,因而能采用较大的矢量维数以达到较佳的数据压缩率。该方法允许任意指定量化器的级数、矢量维数、每采样码数或传输速率,并按任意指定的数目生成初始码本并将它们优化到局部最优点,然后从中挑出最佳者。该方法中误差计算为一独立模块,可通过更换不同模块来实现以不同误差测量准则来优化码本。由于该方法能方便地设计出具有各种不同参数的多级矢量量化器,故能对量化器级数、矢量维数、每采样码数及不同误差测量方法与量化器性能的关系作有效地研究模拟。本文还给出用该法设计的传输速率为16kbit/s及9.6kbit/s两个多级语音波形矢量量化器的模拟结果。

A method is presented to design general Multisiage VQ which can effectively subdue the exponentially growing computational complexity and storage require-ment with the dimension of vector,so that larger dimension of vector can be used to obtain better data compression.This method permits to assign the num-ber of the VQ′s stage,the dimension of the vector,bits per sample or transmission rate,and the number of initial codebook at will.Then these initial codebooks are optimized locally,and the best one of them is chosen.Distortion computation is an independent module,so various distortion measurements can be used to optimize the codebooks simply by altering corresponding module.This method makes it easy to design Multistage VQ with various parameters,and to study and simulate VQ′s performance with the number of stage,dimension of vector,bits per sam-ple and distortion measurement.Simulation results of two speech waveform multi-stage VQs are given which are designed in this method with transmission rate 16 kbits/sec and 9.6 kbits/sec.