数字语音信号LPC分析中的S_w(n)及其选择

Sw(n) IN DIGIT SPEECH SIGNAL LPC ANALYSIS AND SELECTION

本文对数字语音信号s(n)在语音识别过程中的信号预处理阶段使用加窗函数w(n),从引入误差到LPC分析的实用性算法进行了分析,对使用w(n)的必要性和选择进行了讨论。从实践中证实了当语音样点数N与LPC预测阶数P在短时分析中达到N(?)P时,使用s_w(n)及LPC正则方程的高效递推算法较使用s(n)时的算法在维吾尔语语音识别中达到很好的识别效果,提高识别率达3～4个百分点。识别系统实时性有很大的提高。

This paper uses w(n) in the digit speech signal pretreatment in the speech recognition process makes analysis from the introduction of error to practical calculating method of LPC analysis and discusses the necessity of using w(n) and selection. It has been proved that in the uygur speech recognition, the use of Sw(n) and effective recurrence calculation method of LPC authentic equation has a better result than that of the calculation of S(n). The recognition rate has increased 3%-4%,and the real-time property of the recognition system has been greatly improved.