二维哈达玛(Hadamarda)变换在语音识别中的应用

An Application of the Two-Dimensional Hadamarda Transform for Speech Recognition

本文介绍应用二维快速哈达玛变换(FHT)提取语音特征的识别系统,该系统以16个带通滤波器组成的语谱分析电路作为硬件依托,用APPLE-II微型计算机作为语音处理和判决机器,用二维FHT来压缩语音特征量,选用最小距离分类准则作为识别的决策函数。由于FHT的运算速度极快,该系统对中型字库可做到实时识别。系统单字和词组识别率达95％—98％。

In this paper,a speech recognition system is introduced,in which the two-dimensional Fast Hadamard Transform (FHT) is used to extract the speech feature. A Filter band of 16 channels is adopted to prepro-cess the speech signal. After A/D, the digital speech spectrum is entered to APPLE-Ⅱ microcomputer to further processing. The 2-dimensional FHT is mainly used to compress the speech information, which has to be stored in the computer. The nearest neighbour rule is adopted to determine the recognition result. The system can reco(?)nize a moderate-size vocabulary in real-time, this is due to the high speed of the FHT. The correct recognition rate for a vo cabulary of words and phrases is between 95% to 98%.