基于码本共享算法的分模式多级矢量量化  

Mode based multi-stage vector quantization with shared codebooks

作　　者：魏旋[1] 计哲[1] 崔慧娟[1] 唐昆[1] 

机构地区：[1]清华大学电子工程系,微波与数字通信国家重点实验室,北京100084

出　　处：《清华大学学报（自然科学版）》2011年第1期131-134,共4页Journal of Tsinghua University(Science and Technology)

基　　金：国家自然科学基金资助项目(60572081)

摘　　要：为了在存储量受限的情况下尽可能提高线性预测编码(linear predictive coding,LPC)系数量化性能,提出了一种基于码本共享算法的分模式多级矢量量化(multi-stagevector quantization,MSVQ)算法。由于LPC参数的分布与清浊音(unvoiced/voiced,U/V)参数相关,该算法对不同U/V对应的LPC参数进行不同量化,然后利用码本共享算法减少存储量需求。实验表明:在相同码率的情况下,该算法较MSVQ平均谱失真(spectrum distortion,SD)降低3.2%,码本大小增加26.7%;较分模式量化(mode-basedquantization,MBQ)平均谱失真升高3.6%,但是码本尺寸下降了92.1%。该算法是MSVQ与MBQ算法的一种折衷,在增加少量存储量的情况下提高了LPC系数的量化性能。A mode based multi-stage vector quantization（MSVQ） algorithm was developed to more efficiently quantize linear predictive coding（LPC） coefficients with constrained storage.Since the distributions of the LPC parameters are related to the unvoiced/voiced（U/V） parameters,the LPC parameters were quantized according to their U/V parameters,with a codebook sharing algorithm used to reduce the storage requirements.Tests show that the average spectrum distortion（SD） of the quantization algorithm is reduced by 3.2% with the storage is increased by 26.7% compared with the MSVQ algorithm,and the average SD increases by 3.6% and the storage decreases by 92.1% when compared with the mode-based quantization（MBQ） algorithm.The current algorithm is a compromise between the MSVQ and the MBQ algorithms to improve the LPC quantization performance with only a small storage increase.

关 键 词：语音编码 低速率 矢量量化 分模式量化 码本共享 

分 类 号：TN912.32[电子电信—通信与信息系统]