基于递推最小二乘滤波器的语音增强

根据语音信号短时预测特性和短时准周期性，以当前采样前若干点和前若干周期的采样为观测样本，采用横向滤波器结构，构造基于递推最小二乘算法的自适应滤波器。该算法无需先验知识和参考信道，且对平稳或非平稳噪声均适用。实验证明，信噪比有明显提高，且增强后语音信号LPC倒谱与纯净语音LPC倒谱的欧氏距离也比增强前有了明显减小。

A STUDY OF SPEECH ENHANCEMENT BASED ON RLS FILTER

A study of speech enhancement based on adaptive filter with RLS algorithm and transverse architecture is presented in this paper. According to the characters of speech signal, i. e. the short time prediction and the quasi - periodicity sampling taken at the points just before the present one and several periods past are used as observation samples.This algorithm doesn't need priori knowledge or reference channel,and it can be used in the stationary or nonstationary noise environment. Experiment shows it could improve both signal - to - noise ratio and the Eucliden distance of LPC cepstral obviously.