多矢量多模式量化中码本尺寸联合优化算法

Joint optimization algorithm of codebook size in multi-parameter,multi-mode quantization

为了在基于模式的矢量量化中得到更好的量化效果,提出了一种有效的各模式码本尺寸分配算法以逼近全局最优。该算法对于各个模式分别计算每一种码本尺寸所带来的量化失真,在所有模式码本尺寸之和受限的条件下,寻找所有量化失真之和最小的码本尺寸分配方案。结果表明:该算法在运算量和存储量基本不变的情况下,可以有效地降低量化失真,提高量化精度,合成语音平均意见得分(MOS)提高0.02左右。

An efficient algorithm was developed to optimize codebook sizes to improve vector quantization based on different modes. This algorithm first selects several codebook sizes for each mode and then calculates the quantization distortion of each codebook size to find a codebook size allocation with the least quantization distortion for a given sum of all the codebook. Tests show that the method effectively reduces the quantization distortion in the case of invariable operation and storage and increases the mean opinion score （MOS） of the synthesized speech by 0. 02.