This paper describes a fast word pre-selection method for continuous speech recognition. It is essential to implement a bottom-up word pre-selection method to a continuous speech recognition system because the time for recognition grows as dictionary size increases. In this paper, a new word pre-selection method, "the extended redundant hash addressing method" is proposed. This method is an extension of the redundant hash addressing method to word spotting. Moreover, word spotting accuracy is improved using information of phoneme recognition error. Word spotting experiments showed that the following two results: the proposed method was about five times faster than the word spotting using the continuous DP matching and the word spotting accuracy of the proposed method was comparable to that of the continuous DP matching.
Cite as: Ito, A., Makino, S. (1992) Word pre-selection using a redundant hash addressing method for continuous speech recognition. Proc. 2nd International Conference on Spoken Language Processing (ICSLP 1992), 309-312, doi: 10.21437/ICSLP.1992-68
@inproceedings{ito92_icslp,
author={Akinori Ito and Shozo Makino},
title={{Word pre-selection using a redundant hash addressing method for continuous speech recognition}},
year=1992,
booktitle={Proc. 2nd International Conference on Spoken Language Processing (ICSLP 1992)},
pages={309--312},
doi={10.21437/ICSLP.1992-68}
}