一种自适应线谱对滤波器的新算法及在线性预测中的应用

A New Algorithm of Adaptive Line Spectral Pair Filter and its Application on the Linear Prediction

本文从自适应格型滤波器导出自适应线谱对（ＬＳＰ）滤波器，使用最小均方（ＬＭＳ）型自适应算法逐级更新计算线谱对系数。实验表明。该算法与其它算法比较，具有更高的收敛率和较低的失调。用该算法计算得到的ＬＳＰ系数进行语音线性预测合成，获得比使用ＰＡＲＣＯＲ系数更好的效果。

In this paper, an adaptive Line Spectral Pair (LSP) filter is derived from an adaptive lattice filter. A lesst-mean-square(LSP) type adaptive algorithm is used to calculate directly the line spectral pair coefficients on a stage-by-stage basis. Experimental resuets show that the algorithm has higher convergence rate and lower misadjustment as compared with the other algorithm.Using LSP coefficients calculated by the algorithm, speech linear predictive synthesis work has been carried out, and better results have been obtained as compared with that using PARCOR coefficients.