针对Mel频率倒谱系数(Mel frequency cepstralcoefficient,MFCC)特征的说话人确认系统在干净语音环境下具有很高识别率但在噪音环境下识别率急剧下降的缺点,构建了基于感知对数面积比系数(perceptual log area ratio,PLAR)特征的说话人确认系统,并对该系统的噪音鲁棒性进行研究。结果表明:PLAR特征具有较强的噪音鲁棒性。将PLAR与MFCC进行特征域和分数域的融合,利用两者之间存在着的互补性,可有效提高说话人确认系统的识别性能。 Aiming at the shortcomings that the speaker recognition system with Mel frequency cepstralcoefficient (MFCC) feature has a high recognition rate in clean speech environment but the recognition rate drops sharply in noisy environments, a new algorithm based on perceived logarithmic area ratio (perceptual log area ratio, PLAR) feature of the speaker verification system, and study the system noise robustness. The results show that PLAR features have strong noise robustness. The combination of feature domain and fractional domain of PLAR and MFCC makes use of the complementarity between the two, which can effectively improve the recognition performance of speaker verification system.