Word spotting speech recognition device, word spotting speech recognition method, word spotting speech recognition program

摘要

PROBLEM TO BE SOLVED: To solve a problem in which when a matching process for a speech is performed by using a model for a keyword not included in the speech, normalization using an optimum path decreases in normalization cumulative distance and a wrong keyword may be outputted oftener as recognition results.;SOLUTION: A word spotting speech recognition apparatus which normalizes a cumulative distance has a feature parameter generation part (4) which generates a feature parameter from an inputted speech, a keyword model generation part (7), a keyword viterbi calculation part (12) which calculates the cumulative distance while outputting state transition information, a word length calculation part (14) which adds a word length according to the input of the state transition information, but does not add the word length until a next state is entered when a self-transition frequency exceeds a self-transition frequency threshold, and a cumulative distance normalization part (15) which normalizes the cumulative distance from the cumulative distance and word length.;COPYRIGHT: (C)2004,JPO