基于单边自相关线性预测的语音去噪声算法

提出了一种在自相关域上，以相关函数值为参数，利用单边自相关序列的线性预测误差去除语音中加性噪声的方法。该方法首先对含噪语音进行单边自相关处理，以语音信号的单边自相关序列替代语音信号序列，进而对该序列进行线性预测分析后，获得线性预测分析系数，并求得线性预测误差。根据误差能量与信号能量的比例关系，确定减因子u，从含噪语音中根据减因子u的大小减去预测误差，即可抑制噪声误差能量。实验表明；上述方法在低信噪比时，仍能较好地保留语音信号的频谱结构，使音质不至于下降。

Elimilating Speech Noise Algorithm Based on One-sided Autocorrelation Linear Prediction

This paper proposed a new method which can reduce white noise using correlation functions of one-sided autocorrelation and linear prediction residual in the autocorrelation domain. When one-sided autocorrelation sequence is obtained, speech sequence is replaced by one-sided autocorrelation sequence. Then linear prediction coefficients and linear prediction residual are calculated. Based on it, a method of speech denoising is proposed that the noise residual is subtracted from the noisy speech signal in the proportion of subtracting coefficient u, which is gained from proportion of residual and signal energy. Experimental results show that the structure of the speech spectrum is retained well and quality of the speech is not reduced in lower SNR yet.