在矢量量化中，码书起决定性的作用，它决定了量化的性能；一般采用ＬＢＧ 算法生成码书，其中一个关键的技术就是初始码书的选取，通常认为分裂法效果显著．该文引入贪婪树生长算法来设计初始码书，得到了两个优于分裂法的初始码书设计算法，减少了整个码书训练的运算时间，提高了码书的性能． In vector quantization, the codebook plays a decisive role, which determines the performance of quantization; LBG algorithm is generally used to generate codebooks, one of the key technologies is the selection of the initial codebook, usually considered the splitting effect is significant. In this paper, the algorithm of greedy tree growth is introduced to design the initial codebook, and two initial codebook design algorithms are obtained, which are better than the splitting method, which reduces the computing time of the whole codebook training and improves the performance of the codebook.