A General Bayesian Model for Speaker Verification. Issue 6 (1st November 2016)

• Record Type: Journal Article
• Authors: Xu, Yunfei; Yang, Hai; Yang, Lin; Zhou, Ruohua; Yan, Yonghong
• Is Part Of: Chinese journal of electronics. Volume 25:Issue 6(2016)
• Page Start: 1045
• Page End: 1051
• Publication Date: 2016-11-01
• Subjects: Bayesian model -- Speaker verification -- IVector; Bayes methods -- expectation‐maximisation algorithm -- Gaussian processes -- speaker recognition; mismatched train‐test channel conditions -- phone‐call conditions -- Gaussian probabilistic linear discriminant analysis -- SRE -- speaker recognition evaluation -- NIST -- National institute of standards and technology -- detection cost function -- redundant information -- factor analysis technique -- speaker verification decisions -- expectation‐maximization algorithm -- generative probability model -- general Bayesian model; Electronics -- Periodicals; Electronics -- China -- Periodicals; Electronics; China; Periodicals; 621.38105
• DOI: 10.1049/cje.2016.08.024
• Languages: English
• ISSNs: 1022-4653

Abstract

Abstract : This paper presents a general Bayesian model for speaker verification tasks. It is a generative probability model. Due to its simple analytical property, a computationally efficient expectation‐maximization algorithm can be derived to obtain the model parameters. A closedform solution, which allows the scalable size of enrollment set, is given in a full Bayesian way for making speaker verification decisions. Factor analysis technique is employed to model the speaker‐specific components, then the redundant information in this model will be dropped. Experimental results are evaluated by both equal error rate and minimum detection cost function. The proposed approach shows promising results on the National institute of standards and technology (NIST) Speaker recognition evaluation (SRE) 2010 extended and 2012 core tasks. Significant improvement is obtained when comparing with Gaussian probabilistic linear discriminant analysis, especially under phone‐call conditions and mismatched train‐test channel conditions. Contrast experimental results with other popular generative probability models are also presented in this paper.