Reliable word recognition from degraded speech signals requires the incorporation of robust feature extraction as well as noise reduction into the system. This paper is focused on the comparison of different feature extraction methods with respect to different types of distortions. The experiments were made to tackle the problems of additive and convolutional noise as well as with the speaker-stress related Lombard effect, and are part of our ongoing work in the area of robust speech recognition.
Cite as: Eckhardt, H., Trompf, M., Angleys, G., Hackbarth, H. (1992) Robust signal preprocessing for word recognition in noisy environment. Proc. ETRW on Speech Processing in Adverse Conditions, 85-88
@inproceedings{eckhardt92_spac,
author={H. Eckhardt and M. Trompf and G. Angleys and Heidi Hackbarth},
title={{Robust signal preprocessing for word recognition in noisy environment}},
year=1992,
booktitle={Proc. ETRW on Speech Processing in Adverse Conditions},
pages={85--88}
}