In this paper, we propose a new noise compensation method based on the eigenvoice framework in feature space to reduce the mismatch between training and testing environments. In this method, the difference between clean and noisy environments is represented by the linear combination of K eigenvectors that represent the variation among environments. Since how to construct the noisy models is crucial for the performance of the proposed method, we introduce two methods for constructing noisy models : one based on MAP adaptation method and the other using stereo DB. In experiments using Aurora 2 DB, we obtained 44.9% relative improvement with eigen-environment method in comparison with baseline system. Especially, in clean condition training mode, our proposed method yielded 67.4% relative improvement.
Cite as: Song, H.J., Kim, H.S. (2005) Eigen-environment based noise compensation method for robust speech recognition. Proc. Interspeech 2005, 981-984, doi: 10.21437/Interspeech.2005-234
@inproceedings{song05_interspeech,
author={Hwa Jeon Song and Hyung Soon Kim},
title={{Eigen-environment based noise compensation method for robust speech recognition}},
year=2005,
booktitle={Proc. Interspeech 2005},
pages={981--984},
doi={10.21437/Interspeech.2005-234}
}