论文部分内容阅读
本文研究了一种新型的语音多序列激励线性预测(MSLPC)模型。它将激励用有限个长时规则序列表示。MSLPC激励模型中的长时特征使它能较好地利用语音浊音中的长时多余度,并能推出快速搜索算法。 本文在介绍了MSLPC模型的基本概念后,重点讨论了MSLPC激励参数的块模式最大互关快速搜索算法,以及激励序列幅度矢量的两种乘积码矢量量化(VQ)的设计。实验结果表明,MSLPC模型能用于实现在2.4 kb/s至4.8 kb/s速率段的低速率语音编码。
In this paper, we study a new speech multi-series excitation linear prediction (MSLPC) model. It will be motivated by a finite number of long-term rule sequences. Long-term features in the MSLPC motivation model make it possible to make good use of long-term redundancy in speech voices and to introduce fast search algorithms. After introducing the basic concepts of MSLPC model, this paper focuses on the fast search algorithm of block mode maximum cross-correlation of MSLPC excitation parameters and the design of two kinds of product code vector quantization (VQ) of excitation sequence amplitude vector. Experimental results show that the MSLPC model can be used to implement low-rate speech coding at 2.4 kb / s to 4.8 kb / s.