Speech recognition apparatus with tree style language model

摘要

PURPOSE: A continuous voice recognition device having a language model of a tree structure is provided to decrease a storage capacity, and to increase the recognition speed by expressing multi-step language model as one tree structure. CONSTITUTION: A tree has a root, and each word used at a system is located at the root. One word is expressed every node of the tree structure. If the gap between the root of a word w1 and an optional node is d, and if a word sequence is h=w1,w2,w3,...,wd, a probability at the node of depth d is p(w1/wd,wd-1,..,w2).