改进的跨语种说话人确认方法的研究

Research of improved cross-lingual speaker verification method

提出了一种基于改进的语音融合特征和GMM模型相结合的跨语种说话人确认方法.首先,采用Teager能量算子提取语音中的浊音段,消除与说话人声道特征无关的静音段和清音段.其次,提取基音周期参数,并与16维的MFCC参数融合形成本文的语音融合特征.最后,将本文方法与文献[9]的方法分别进行了单语种和跨语种的说话人确认对比实验,实验结果表明本文方法识别准确率和平均判别时间均优于文献[9]的方法,证明本文提出的方法有效,可用于跨语种的说话人确认应用领域.

This paper presents a cross-lingual speaker verification method based on improved speech fusion feature and GMM model.First,the Teager energy operator is used to extract voiced clips in speech,eliminating mute and unvoiced clips that are independent of speaker＇s vocal tract.Secondly,pitch period parameters are extracted and fused with 16-dimensional MFCC parameters to form speech fusion feature.Finally,experimental results show that the accuracy and average discriminant time of this method are better than that of reference [9],which proves that the method proposed in this paper is valid and available in cross-lingual speaker verification applications.