针对Mel频率倒谱系数(Mel frequency cepstral coefficient,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.
Noise-robustness of speaker verification based on the perceptual log area ratio
Speaker verification systems based on Mel frequency cepstral coefficients (MFCCs) have higher results in clean conditions,but the results are sharply worse in noisy environments.This paper presents a noise-robust speaker verification system based on perceptual log area ratio (PLAR) feature extraction.The results show that the PLAR is more robust to noise than the MFCCs.PLARs and MFCCs are complimentary with fusion of these two features in the feature domain and the score domain effectively improving speaker verification auuracy in noisy conditions.
speaker verificationperceptual log area ratio (PLAR)robustnessfusion
NETL
NSTL
万方数据
维普
