Sixth European Conference on Speech Communication and Technology

Budapest, Hungary
September 5-9, 1999

A DCT-based Fast Enhancement Technique for Robust Speech Recognition in Automobile Usage

Jun Huang (1), Yunxin Zhao (2), Stephen Levinson (1)

(1) Beckman Institute and Dept. of ECE, University of Illinois, Urbana, IL, USA
(2) Department of CECS, University of Missouri, Columbia, MO, USA

In this paper, a fast computational method is proposed to approximate the Karhunen-Lo_eve transform (KLT) in signal-subspace-based speech enhancement algorithm. The discrete cosine transform (DCT) is shown to be a good approximation of KLT for the covariance matrix of the autoregressive process of order p (AR(p)). A fast algorithm which reduces the computation of eigenvalues of an N _ N symmetric Toeplitz matrix from O(N 3) inKLT to O(N 2 ) is developed. Experiment results demonstrate that the performance of the fast algorithm is very close to that of the KLT-based method in robust speech recognition in car environment while significantly reduces the computation time. An acoustic normalization scheme is also found to be usful to compensate the mismatch between the training and test conditions and thus further improves the recognition performance.

Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Huang, Jun / Zhao, Yunxin / Levinson, Stephen (1999): "A DCT-based fast enhancement technique for robust speech recognition in automobile usage", In EUROSPEECH'99, 1947-1950.