Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Language Recognition Using Time-Frequency Principal Component Analysis and Acoustic Modeling

Michel Dutat (1,2), Ivan Magrin-Chagnolleau (3), Frédéric Bimbot (3)

(1) LSCP / CNRS, Paris, France
(2) ENST / CNRS, Dépt. Signal et Image, Paris, France
(3) IRISA (CNRS & INRIA), Campus universitaire de Beaulieu, Rennes, France

Time-Frequency Principal Component (TFPC) is a speech parameterization technique based on a principal component analysis applied to acoustic feature parameters augmented by their time context. In this paper, we investigate on the performance of TFPC in the framework of automatic language recognition. In our experiments, identification rate is improved compared to the use of the conventional cepstral coefficients augmented by their Δ coefficients.

Full Paper

Bibliographic reference.  Dutat, Michel / Magrin-Chagnolleau, Ivan / Bimbot, Frédéric (2000): "Language recognition using time-frequency principal component analysis and acoustic modeling", In ICSLP-2000, vol.2, 230-233.