语音信号线性预测参数的韧性估计

Robust Estimate for Linear Prediction Parameters of Speech Signal

使用牛顿算法的韧性Ｍ估计方法来估计语音信号的线性预测参数．该方法考虑了语音信号模型浊音激励源的非高斯特性，它能消除浊音激励中远离值的影响．使用合成语音信号数据的实验表明，与常规的线性预测方法比较，所提出的方法能给出更有效和更小偏倚的估计．

The robust M-estimate procedure in a Newton-type algorithm is used to estimate the LP parameters of an autoregressive (AR) speech signal model. The procedure takes into account the nonGaussian nature of the excitation for voiced sound and can eliminate the influence from the outliers. Experiments with synthesized speech show that the estimate procedure proposed gives more efficient and less biased estimetes than those by conventional LP methods.