Maximum Likelihood Discriminant Feature for Text-Independent Speaker Verification

Author

Liu, Qingsong ; Dai, Beiqian

Author_Institution

Dept. of Electron. Sci. & Technol., Univ. of Sci. & Technol. of China, Hefei, China

fYear

2009

fDate

17-19 Oct. 2009

Firstpage

1

Lastpage

4

Abstract

Feature extraction is an essential first step in speaker verification applications. In addition to static features extracted from each frame of speech data, it is beneficial to use dynamic features that use information from neighboring frames. In this paper a new feature estimation method based on maximum likelihood discriminant analysis is presented. We compare it to traditional MFCC features in a NIST 2006 SRE core task. Experiments show that the proposed scheme provides more discriminative feature vectors. The features obtained with the new estimation method show a 10% -15% relative improvement in EER and MinDCF over traditional MFCC features.

Keywords

feature extraction; maximum likelihood estimation; speech recognition; NIST 2006 SRE core task; feature estimation method; feature extraction; feature vector; maximum likelihood discriminant feature; text independent speaker verification; Cepstral analysis; Covariance matrix; Data mining; Decorrelation; Linear discriminant analysis; Maximum likelihood estimation; Mel frequency cepstral coefficient; Performance analysis; Speech; Vectors;

fLanguage

English

Publisher

ieee

Conference_Titel

Image and Signal Processing, 2009. CISP '09. 2nd International Congress on

Conference_Location

Tianjin

Print_ISBN

978-1-4244-4129-7

Electronic_ISBN

978-1-4244-4131-0

Type

conf

DOI

10.1109/CISP.2009.5301537

Filename

5301537