一种基于LPC残差的语音激励脉冲逼近方法

A New Method for Approaching Speech Exciting Pulse Based on LPC Residual

针对某些语音LPC(L inear Pred iction Cod ing)分析的缺陷提出一种改进算法。该算法重点研究经典LPC分析后基音激励方向向下的语音,对这种浊音LPC残差进行后滤波以取代预增强的方法使其逼近语音激励。该算法将传统LPC分析中的声道模型和声门模型分开考虑,既避免了引入ARMA模型难以计算的缺陷,同时又显著的抵消了声门模型中极点的影响。实验表明,该算法对经典LPC分析后基音激励方向反向的语音,改善效果明显,残差的方向性与理论分析更加吻合。最后将该方案应用于语音水印的研究中,具有一定实用性。

This paper proposes a novel algorithm to overcome the disadvantage of LPC （ Linear Prediction Coding）. Firstly we analyze the speech with reversed polarity and present the idea of post-filtering of the LPC residual of these voiced speech frames. The proposed algorithm reduces computational complexity of ARMA （ Autoregressive Moving Average） model and the effect of pole-points in glottis model. Experimental results show that this new algorithm obviously improves the performance of LPC. Finally, we describe the usage of this method in audio information hiding.