在语音压缩编码中，矢量量化占有很大比重和计算量．码本的好坏与方法的选择有很大的关系．笔者介绍了一种方法，将神经网络和隐含马尔可夫过程应用于矢量量化，从而产生了更为优良的码本． In speech compression coding, vector quantization occupy a large proportion and amount of computation. Good or bad codebook and the choice of method has a great relationship. The author introduces a method that applies neural networks and implicit Markov processes to vector quantization, resulting in a much better codebook.