基于遗传K-均值算法的LSF参数码书设计

LSF Parameter Vector Quantization CodeBook Design Based On Genetic k-Means Algorithm

文章提出了一种基于遗传K-均值算法的线谱频率参数(LSF)码书设计方法。该方法使用K-均值算法定义遗传操作中的交叉操作,采用训练矢量随机替换码书码字的方法进行变异操作。实验结果表明:该方法是有效的,在相同的码书尺寸下,用该量化器设计的LSF码书较传统分裂式LBG算法(初始码书由分裂法产生)设计的码书质量有所提高。

A Line Spectrum Frequency （LSF） parameter codebook design method based on genetic k-means algorithm is introduced. By using k-means algorithm as a novel genetic crossover operator and the method of replacing the codebook＇ s codes by the training vectors randomly as the genetic mutation operator, the experimental result shows that it is effective and the code book generated by this method is better than the one generated by the traditional splitting LBG method with the same codebook size.