Abstract: The traditional algorithm LBG is often adopted in designing of codebooks of speech parameter LSF.Since LBG easily traps in local optima，a new algorithm is represented in this paper combining differential evolution with LBG.The hybrid algorithm DELBG introduces a weighed factor，and takes advantage that the chromosome of the factor is very short and is easily to be optimized by DE to alter the search path and then makes LBG escape from local optima as possible as it can.DELBG also avoids the problem that DE’s search space is too large before arriving at the global optimum when DE is solely used to design LSF codebooks.Experiments show the represented algorithm DELBG better realizes global optimization compared with LBG.

Key words: vector quantization, design of codebook, LBG, differential evolution, line spectrum frequency

摘要： 线谱对（LSF）的码本设计常采用经典的LBG算法，由于该算法容易陷入局部最优，提出了一种结合差分进化（DE）和LBG算法的混合算法（DELBG），通过引入一加权因子，利用其染色体表示很短、易于被差分进化优化的特点不断改变LBG算法的搜索路径，最大可能地使之逃离局部最优，同时也避免了差分进化在单独设计LSF码本中达到全局最优解前搜索空间过大的问题。实验表明提出的DELBG算法相比LBG算法较好地实现了全局最优。

关键词: 矢量量化, 码本设计, LBG, 差分进化, 线谱对