论文部分内容阅读
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。
In order to get a better quantization result in pattern-based vector quantization, an effective size distribution algorithm for each mode code is proposed to approximate the global optimum. The algorithm separately calculates the quantization distortion caused by each codebook size for each mode, and searches for the codebook size allocation scheme with the smallest sum of quantization distortion under the condition that the sum of the size of all the codebook modes is limited. The results show that this algorithm can effectively reduce the quantization distortion and improve the quantization accuracy under the condition of the same amount of operation and memory. The MOS of the synthesized speech improves by about 0.02.