语音动力学系统的神经网络建模方法研究

RESEARCH ON MODELLING VOICE NON-LINEAR DYNAMIC SYSTEMS

摘要: 人工神经网络（ANN）方法是非线性动力学系统建模的有效方法．本文针对多层ANN结构，运用递推预报误差（RPE）算法对离散非线性动力学系统进行了建模研究，并将之运用于语音非线性动力学系统的动态建模，估计出了语音非线性动力学系统稳态吸引子的维数，为了解语音和实用化的语音识别提供了良好的基础．

Abstract: This paper presents, the use of recursive prediction error algorithm(RPE) of training multilayered artificial neural neural(ANN) networks in discrete time non-linear dynamic system identification. As an example, the nonlinear voice dynamic system is modelled based on RPE algorithm. The results provide a good foundation to understand and identify voice model.