This paper investigates a projection-based likelihood measure that improves speech recognition performance in noisy environment. The projection-based likelihood measure is modified to give the weighting and projection effect and to reduce computational complexity. It is evaluated in sub-model based word recognition using semi-continuous hidden Markov model with speaker independent mode. Experimental results using proposed measure are reported for several performance factors: additive noise and noisy channel environment, various noise signals, and combination with other compensation method. In various noisy environments, performance improvements were achieved compared to the previously existing methods.
Cite as: Shin, W.-H., Kim, W.-G., Lee, C., Cha, I.-W. (1998) Speech recognition in noisy environment using weighted projection-based likelihood measure. Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998), paper 0434, doi: 10.21437/ICSLP.1998-339
@inproceedings{shin98_icslp,
author={Won-Ho Shin and Weon-Goo Kim and Chungyong Lee and Il-Whan Cha},
title={{Speech recognition in noisy environment using weighted projection-based likelihood measure}},
year=1998,
booktitle={Proc. 5th International Conference on Spoken Language Processing (ICSLP 1998)},
pages={paper 0434},
doi={10.21437/ICSLP.1998-339}
}