论文部分内容阅读
随机码本是码本激励线性预测CELP声码器的一个重要组成部分,其码本结构及其相应的搜索算法直接影响着语音编译码系统的语音质量和实时实现。本文给出了一个结构化随机码本──“线性移位”高斯码本的一组双向递归计算公式,由此提出了一种单向递归搜索算法,从而有效地降低了随机码本的搜索复杂度,使搜索速度较之参考文献给出的原随机码本搜索法提高1.76倍。为进一步减小码本搜索的运算量,本文还引入了局部搜索的次优化准则,提出了双向递归搜索算法,在保持相同合成音质的条件下,使码本搜索速度再次提高一倍,即可以使速度提高为原随机码本搜索法的3.5倍。
Random codebook is an important part of CELP vocoder based on Code Excited Linear Prediction. Its codebook structure and its corresponding search algorithm directly affect speech quality and real-time implementation of speech codec system. In this paper, we present a set of bidirectional recursive formulas for the calculation of a “random” linearized Gaussian codebook, and propose a one-way recursive search algorithm, which can effectively reduce the random codebook search Complexity, the search speed compared with the reference given by the original random codebook search method to improve 1.76 times. In order to further reduce the computational complexity of codebook search, this paper also introduces the sub-optimization criterion of local search and proposes a two-way recursive search algorithm, which can double the speed of codebook search again while maintaining the same synthesized sound quality 3.5 times faster than the original random codebook search method.