Cross-word search algorithm based on two-layer lexical tree for speech recognition
Jing Li · 2003
A cr oss-word search algorithm was developed based on a two-layer lexical tree usin g context-dependent acoustic models for efficient continuous speech recognition . The algorithm uses the two-layer lexcial tree to maintain the search network to keep the search space from becoming more complicated during the cross-word s earch. The search algorithm can then use context-dependent models on the word b oundary for good performance. Experimental results show that the search algorith m can work in real-time with 60% less word errors compared with the intra-word search algorithm, which illustrates the efficiency of the cross-word search al gorithm based on the two-layer lexical tree.