G.723.1语音编码算法闭环基音搜索的低复杂度算法

Low Complexity Algorithm of Close-Loop Pitch Search in G.723.1Speech Codec

由于ITU-TG.723.1语音编码算法具有较高的算法复杂度,故而在应用与实现时受到了很多的限制。该文提出一种低复杂度闭环基音搜索算法,该算法仍以5阶基音预测器为基础,但在求取5个基音预测增益时不是采用原算法中对20维矢量码本进行搜索的方法,而是利用这个20维矢量组成一个Wiener-Hopf方程,并利用语音的短时平稳特性将该方程简化为一个Toeplitz线性代数方程组,方程组的解就是所求的基音预测增益。对该增益进行5维码本矢量量化,从而用5维矢量码本搜索代替了原来的20维矢量码本搜索。这样使闭环基音搜索部分的运算量降低了一半,语音质量只有略微下降,同时与G.723.1算法码流兼容。

The application and implementation of ITU-T G.723.1speech codec is narrowly limited because of the high computational complexity.A new close-loop pitch search algorithm is presented in this paper.This algorithm is still based on the5-order pitch predictor.A Wiener-Hopf equation is established from the20-dimention vector which is used in the20-dimention codebook search in G.723.1algorithm.The equation solution is exactly the optimal pitch gain and is vector quantized with a5-dimention vector codebook.The synthesized speech quality degrades slightly and the code is compatible with the G.723.1algorithm while the computational complexity decreases near a half.