4th International Conference on Spoken Language Processing
Philadelphia, PA, USA
In this paper, a structural search using a word-node graph is proposed to speed up the isolated word recognition based on hidden markov models (HMMs). We define a distance measure for comparing pairs of words, and construct a graph keeping the structure of word distribution. Based on this graph, the number of words to be examined are restricted. Experiments show that the search complexity is considerably reduced with little degradation of the recognition accuracy.
Bibliographic reference. Minamino, Katsuki (1996): "Large vocabulary word recognition based on a graph-structured dictionary", In ICSLP-1996, 2123-2126.