中文摘要：

说话人识别中的首要问题是从语音信号中提取能唯一表现说话人个性特征的有效而稳定可靠的特征参数。把感知加权技术应用到Mel倒谱分析中，通过对基于心理声学模型计算得到的信号掩蔽比插值获得权重函数，并将权重函数应用到Mel倒谱分析中获得加权Mel倒谱系数（WMCEP），以此为特征进行说话人识别。实验结果表明，WMCEP比MFCC和Mel倒谱系数（MCEP）能更好地逼近说话人的谱包络，在噪声环境下的鲁棒性更好，因此其识别性能要优于MFCC和MCEP。

英文摘要：

The primary issue of speaker recognition is to extract the unique, effective, stable and reliable features that stand for the person- ality of the speaker from speech signals. In this paper we applied the perceptual weighting technology to reel-frequency eepstrum analysis and acquired weighting function from signal-to-mask ratios （SMRS） interpolation which is derived from psychoacoustie model-based calculation, the weighting function was then used in mel-frequency cepstum analysis for obtaining the weighted mel-frequency cepstrum coefficients （ WMCEP）, and take it as the features for recognising the speaker. Experimental results showed that the WMCEP approaches speaker＇ s spectral envelope much better than MFCC and MCEP do, and has better robustness in noisy environment, so that it outperforms the other two in speaker recognition.