说话人识别是语音识别的一种特殊方式，其目的不是识别语音内容，而是识别说话人是谁，即从语音信号中提取个人特征。采用矢量量化（VQ）可避免困难的语音分段问题和时间归整问题，且作为一种数据压缩手段可大大减少系统所需的数据存储量。本文提出了识别特征选取采用复倒谱特征参数和对应用VQ的说话人识别系统改进的一种方法。当用于训练的数据量较小时，复倒谱特征可以得到比较稳定的识别性能。VQ的改进方法避免了说话人识别系统的训练时间与使用时间相差过长从而导致系统的性能明显下降以及若利用自相关函数带来的大量运算。-Speaker Recognition Speech Recognition is a special way, its purpose is not voice recognition, Who identification but said that the voice signal from extracting personal characteristics. Vector quantization (VQ) can avoid the difficulties subparagraph voice to the issues and the whole time, and as a means of data compression system can significantly reduce the required data storage capacity. This paper presents a selection of identifiers employ Cepstrum parameters and the application of VQ speaker recognition system to improve a side France. When training for the amount of data is smaller, rehabilitation Cepstrum be relatively stable recognition performance. VQ improved ways to avoid the speech recognition system of training and the use of the difference in time, resulting in excessive sys