Abstract: The speech production is nonlinear and chaotic.Nonlinear models for speech in the reconstructed phase space is more appropriate for the actual systems than the linear one.Artificial neural network has been widely used in modeling the nonlinear systems.The experiment results indicate that the nonlinear predictor based on RBF network is more accurate than the linear predictor.

摘要： 语音信号的产生过程是非线性的，而且具有混沌性。相对于线性模型，在重构相空间中建立的语音信号模型更接近实际系统，神经网络是建立非线性系统模型的常用工具。实验结果表明：在重构相空间中建立的基于径向基函数神经网络的预测器较线性预测器在性能上有明显提高。