一种语音信号线谱对参数的优化算法

Optimization Design for Efficient Computation of Line Spectrum Pairs

提出了一种计算语音信号线谱对(LSP)参数的新算法.LSP参数可以通过求解一个与线性预测系数(LPC)相关的非线性N阶方程得到.首先利用埃特金迭代法求高阶非线性方程的一个实数根,再采用多项式综合除法降阶,最后运用费拉里算法求解一元四次方程的根,得到LSP系数.理论分析以及基于TMS320C5510 DSP的实验结果表明,文中提出的算法与目前的算法相比,不仅计算精度高、运算量小,完全可以实时处理,而且并可应用于G.729A标准等低码率实时语音传输系统.

This paper proposes a new algorithm APF for calculating line spectrum pairs （LSP） of speech signal. LSP parameters can be obtained by solving an N-order non-linear equation relating to the linear predictive coding （LPC） coefficients. First, one root of the N-order nonlinear equation is derived with the Aitken iterative method. The order is then decreased with polynomial synthesis division. Finally the fourth-order equation is solved using the Ferrari＇ s solution. Theoretical analysis and experiments on TMS320C5510 DSP show that the proposed algorithm has high precision and low computation complexity, meeting the real-time processing requirement. The algorithm can be applied to low-bit real-time speech coding such as G.729A systems.