改进的RBF神经网络非线性预测模型在语音编码中的应用

Application of Nonlinear Prediction Model Based on Improved RBF Neutral Network in Speech Coding System

用一种改进的径向基函数(RBF)神经网络建立非线性预测模型来对语音信号进行处理,在提高预测精度的同时不影响传输码率。这种改进的RBF神经网络具有计算量小,学习速度快,不易陷入局部极小等优点。将此模型应用在ADPCM语音编码系统中进行验证,其恢复的语音质量优于CCITT建议G.721中的ADPCM编码,表明该非线性预测模型具有较高的预测精度,且在语音编码系统中有着很大的实用性。

This paper uses an improved RBF neural network in establishment of a model of nonlinear prediction in order to process speech signals.As a result,the prediction accuracy is improved without any negative influence on the transmission bit rate.The improved RBF neural network features low computational complexity,fast learning capability without easily getting in local minimum.The model is validated in an ADPCM speech coding system.Test results show that the recovery of speech quality is superior to ADPCM coding in G.721 recommended by CCITT.This indicates that the model of nonlinear prediction has higher prediction accuracy and is of practical significance to speech coding systems.