本系统是一个在多媒体PC 上实现的孤立词识别系统, 它提取语音的线性预测系数作为特征参数, 并采用Itaku ra 失真测度计算帧间距离, 在识别上则使用了动态时轴弯曲(DTW ) 进行时间匹配。本系统对一般的DTW 法作了改进, 即通过放宽端点限制以得到更好的语音匹配, 克服了一般DTW 法要求语音首尾严格对齐而造成的弊病, 降低了语音端点检测的精度要求。-This system is implemented on a multimedia PC isolated word recognition system, it extracts the voice of the linear prediction coefficient as the feature parameters, and calculated using Itaku ra frame distortion measure distance, in recognition of the dynamic time is used on the shaft bending (DTW ) for time matching. The system has been improved in general DTW method, namely, by relaxing restrictions in order to get a better endpoint voice match, to overcome the sum of the general DTW law requires strict alignment of voice caused by defects, reducing the speech endpoint detection accuracy.