基于 PRNN神经网络与 FIR子带滤波器 ,本文提出一种新颖的非线性自适应预测滤波器。子带分解采用便于实时实现的 DCT,文中给出了相应的学习算法。对实际语音信号进行的非线性预测结果表明 ,本文所提出的非线性预测滤波器与原来的相比 ,在降低预测误差动态范围与提高收敛速度等方面都有了十分明显的改善。 Based on PRNN neural network and FIR subband filter, this paper presents a novel non-linear adaptive prediction filter. The subband decomposition uses DCT which is easy to realize in real time. The corresponding learning algorithm is given in this paper. The nonlinear prediction of the actual speech signal shows that the nonlinear prediction filter proposed in this paper has obvious improvement compared with the original one in terms of reducing the dynamic range of prediction error and increasing the convergence speed.