摘要： 采用了一种基于逆滤波器原理的方法来识别共振峰参数。用一组逆滤波器来逼近语音信号的短时能谱，通过动态搜索算法可以得到滤波器组中各个滤波器的位置与带宽，从而确定了共振峰的位置和带宽。采用了优化动态搜索的算法，提高了运算速度。把共振峰作为特征参数用于一个小词汇量的汉语语音识别系统中，识别正确率达到了98%。

关键词: 共振峰, 语音识别, 逆滤波器

Abstract: A bank of inverse filters is used to approach the STFT of speech signal. The dynamic programming algorithm searches the optimum locations and bandwidths of the filter bank, so that the location and bandwidth of the filter (also the formant) is specified. The formant and corresponding bandwidth parameters result in a correct recognition rate of 98% in a small vocabulary mandarin speech recognition system.

中图分类号: 

• TN912.34