本文介绍应用线性予测参数推算语音功率谱及共振蜂参数的方法。这是一种计算量小、而且可靠的方法。用所开发的算法,测定了各种姿态下发音的变化,所得的基音频率值证明:正坐时发音的重复性好,其它姿态下变化比较大。这一实验部分地说明了有关识别精度的原因。 This article describes the application of linear predictive parameters to calculate the speech power spectrum and resonant bee parameters. This is a computationally intensive and reliable method. With the developed algorithm, the changes of pronunciations under various gestures were measured. The results of the fundamental frequency showed that the repeatability of pronunciation was good while sitting, and the changes were relatively great under other gestures. This experiment partly explains the reason for the recognition accuracy.