1.引言语音信号在进入语音识别系统之前,通常受到一些诸如语音产生过程、记录过程以及传输过程中产生的通道畸变的影响。例如:麦克风依赖于其类型和位置的不同能明显地影响语音信号的频谱,电话线网络的频率特性也对语音信号产生了频谱畸变的影响。这些由于传输通道而引入的频谱畸变将直接影响到语音信号的短时频谱分析结果,目前所有的语音识别系统中的参数计算,以及相似度测度都直接或间接地依赖于语音信号的短时频谱分析结果, 1. INTRODUCTION Before entering speech recognition systems, speech signals are usually affected by some channels such as speech generation, recording and channel distortion during transmission. For example, depending on the type and location of the microphones, the frequency spectrum of the voice signal can be obviously influenced. The frequency characteristics of the telephone line network also affect the spectral distortion of the voice signal. These spectral distortions introduced by the transmission channel will directly affect the short-term spectrum analysis results of the speech signal. At present, the calculation of the parameters in all speech recognition systems and the similarity measure depend directly or indirectly on the short-time frequency spectrum of the speech signal Analysis results,