本文提出一种新的矢量量化方法压缩语音特征用于孤立字语音识别.新方法借鉴了隐马尔柯夫模型(HMM)中状态的概念来规划模板;用动态规划(Dyna-mic Progamming)的技术优化矢量量化产生的初始码字.新方法使得识别所需的模板库的尺寸明显减小,识别响应时问缩短,而且由于模板的优化,使得系统的正确识别率显著提高. In this paper, a new vector quantization method is proposed to compress speech features for speech recognition in isolated words. The new method uses the concept of state in Hidden Markov Model (HMM) to plan the template. With dynamic programming (Dyna-mic Progamming) The initial codeword is optimized by vector quantization.The new method makes the size of the template library needed for identification significantly reduced and the recognition response time shortened, and the correct recognition rate of the system is significantly improved due to the template optimization.