公开(公告)号：KR1020010064135A

公开(公告)日：2001-07-09

申请号：KR1019990062269

申请日：1999-12-24

Applicant: 한국전자통신연구원

Inventor： 윤병식 ,  강상원 ,  손창용 ,  김형중 ,  이정철

IPC: H03M7/50

CPC classification number: G10L19/06

Abstract: PURPOSE: A fast searching method of a line spectrum pair quantizer according to predictive split vector quantization and predictive split matrix quantization is provided which reduces the size of a code book to be searched using sequence characteristic of line spectrum frequency coefficient in a vector code book searching procedure requiring lots of calculations, to decrease the amount of calculations without attenuating spectrum distortion performance. CONSTITUTION: A criterion is obtained using a target vector changed to have sequence characteristic and an object code vector. A DC component is added to the object code vector and a new code book for five sub-matrices is made. A specific row used for deciding a searching range is selected from a code book reconstructed to have sequence characteristic and code vectors are descending-aligned. The searching range is determined using the sequence characteristic of the target vector having the sequence characteristic and aligned code vectors. The criterion is calculated only within the determined searching range to find an optimal object code vector.

Abstract translation: 目的：提供一种基于预测分裂矢量量化和预测分割矩阵量化的线谱对量化器的快速搜索方法，其使用矢量码本搜索中的线谱频率系数的序列特征来减少要搜索的码本的大小 需要大量计算的程序，减少计算量而不衰减频谱失真性能。 构成：使用改变为具有序列特征的目标向量和目标代码向量来获得标准。 DC分量被添加到目标代码向量中，并且制作了一个用于五个子矩阵的新代码簿。 用于确定搜索范围的特定行从重构为具有序列特征的代码簿中选择，并且代码向量是降序对齐的。 使用具有序列特征和对齐的码矢量的目标矢量的序列特征来确定搜索范围。 仅在所确定的搜索范围内计算标准以找到最佳目标码矢量。

公开(公告)号：KR100324204B1

公开(公告)日：2002-02-16

申请号：KR1019990062269

申请日：1999-12-24

Applicant: 한국전자통신연구원

Inventor： 윤병식 ,  강상원 ,  손창용 ,  김형중 ,  이정철

IPC: H03M7/50

CPC classification number: G10L19/06

Abstract: 본발명은선형스펙트럼쌍(LSP : Line Spectrum Pair) 계수의예측분할행렬양자화(PSMQ : Predictive Split Metrix Quantization) 또는예측분할벡터양자화(PSVQ : Predictive Split Vector Quantization) 방식을사용하는음성부호화기에관한것으로서, 보다상세하게설명하면코드북탐색시요구되는계산량을줄이기위한예측분할벡터양자화및 예측분할행렬양자화방식에의한선스펙트럼쌍양자화기의고속탐색방법에관한것이다. 본발명에따른예측분할벡터양자화및 예측분할행렬양자화방식에선스펙트럼쌍양자화기의고속탐색방법은, 순서성질을갖는목표벡터와에러코드북을구성하고, 순서성질을갖도록재구성된각 코드북은선정된하나의특정열의요소값을근거로전체코드북을내림차순으로정렬한다. 그리고, 각코드북내 정렬된특정열의요소값과목표벡터내 해당열전후의요소값들을순방향및 역방향비교를통해순서성질을만족하는대상코드벡터들의검색영역의시작점과끝점을구한다. 이렇게시작점과끝점이구해지면, 이탐색범위내에서만계산량이많은에러기준값(criterion) E을구하여최적의대상코드벡터를구한다.