Suppressing Phonetic Information in Text-Independent Speaker Identification via LPP

摘要：

How to suppress the phonetic information in speech features and extract the elements only contains speaker characteristic is an urgent problem in textindependent(TI) speaker recognition. Based on the assumption that phonetic information embedded in the lower manifold space of speech features, this paper proposes a construction method to suppress phonetic information in the speech feature space. Using manifold learning method, Locality- Preserving Projection(LPP) to confine most phonetic information to fairly fewer components, which will not participate in the speaker recognition, then the other components with phonetic information suppressed contain purer speaker characteristic, so TI speaker recognition with higher accuracy can obtained via this method. Based on Guassian Mixture Model (GMM), the proposed method can greatly reduce the identification error rate compared with conventional Mel-Frequence Cepstral Coefficient ( MFCC).

关键词： TI speaker recognition manifold learning PCA LPP MFCC GMM

作者: Shaomei Li Lixiong Liu Fucai Chen

作者单位: National Digital Switching System Engineering&Technological R&D Center Zhengzhou.China

会议类型: 国际会议

会议名称: 2011 3rd International Conference on Computer and Network Technology(ICCNT 2011)(2011第三届IEEE计算机与网络技术国际会议)

会议地点: 太原

会议语种:英文

页码: 42-45

在线出版日期: 2011-02-26（万方平台首次上网日期，不代表论文的发表时间）