神经网络非线性预测及语音编码

Non-Linear Prediction Based on ANN and Speech Coding

本文用神经网络非线性模型进行语音信号的处理．通过与各种线性预测算法的结果进行比较表明，前向神经网仅用10个左右的输入抽头可消除语音信号中的短项（相邻样值）和长项（基音）相关性，预测误差小于经典的线性预测系统，这说明所谓的线性基音相关性可表示成语音的短项非线性相关性，在众多的应用领域中就可取消有时难于难确进行的基者估计，本文提出的基于非线性预测的ADPCM语音编码系统。

This paper indicates that non-linear prediction technique should be one of more efficient tools for signal prmessing, because some non-linear correlation exists between adjacent saxnples of various signals. Ex periments based on both linear and non-linear predictions of speech signals are given. Finally, we apply non-lin ear technique to speech coding.