4th International Conference on Spoken Language Processing
Philadelphia, PA, USA
New techniques are developed for the second pass search in our large vocabulary continuous speech recognition system. The merging of recognition hypotheses is proposed in order to linearize the exponential growth of the tree structure in the depth first search. Branching ordering of the first pass word graph and pruning at both word and phone levels are used to further speed up the search. The algorithm has been shown to be effective on the speaker-independent WSJ task.
Bibliographic reference. Li, Z. / Heon, M. / O'Shaughnessy, Douglas (1996): "New developments in the INRS continuous speech recognition system", In ICSLP-1996, 2-5.