Two Schemes for Automatic Speaker Recognition over VOIP

Author

Dan, Qu ; Honggang, Yan ; Hui, Tang ; Bingxi, Wang

Author_Institution

Dept. of Signal Analyzing & Eng., Zhengzhou Inf. Sci. & Technol. Inst., Zhengzhou

Volume

2

fYear

2008

fDate

19-20 Dec. 2008

Firstpage

695

Lastpage

699

Abstract

The paper presents the two schemes of Automatic speaker recognition (ASR) over VOIP including speaker recognition based on decoded codec parameters and compressed packet stream over VoIP. For ASR based on decoded codec parameters, the optimal feature selection using VoIP codec parameters is introduced. For ASR based on compressed bit stream, the paper presents two approaches for compressed speaker recognition respectively based on the Probabilistic Stochastic Histogram algorithm including Vector Quantization Probabilistic Stochastic Histogram (VQPSH) algorithm and Gaussian Mixture Model Probabilistic Stochastic Histogram (GMMPSH). The Two schemes for speaker recognition is test using G.729, G.723.1 6.3K, G723.15.3K compressed bit stream. The experimental results show whether two schemes are very effective for automatic speaker recognition over VOIP.

Keywords

Gaussian processes; Internet telephony; feature extraction; probability; speaker recognition; speech coding; vector quantisation; Gaussian mixture model probabilistic stochastic histogram algorithm; VoIP decoded codec parameter; automatic speaker recognition; optimal feature selection; packet stream compression; vector quantization probabilistic stochastic histogram algorithm; Automatic speech recognition; Bit rate; Codecs; Decoding; Histograms; Internet telephony; Interpolation; Speaker recognition; Stochastic processes; Vector quantization; Automatic; Speaker Recognition; VOIP;

fLanguage

English

Publisher

ieee

Conference_Titel

Computational Intelligence and Industrial Application, 2008. PACIIA '08. Pacific-Asia Workshop on

Conference_Location

Wuhan

Print_ISBN

978-0-7695-3490-9

Type

conf

DOI

10.1109/PACIIA.2008.224

Filename

4756865