摘要 |
PROBLEM TO BE SOLVED: To suppress development of hypotheses according to a plurality of adjacent environments to increase the processing speed of voice recognition, when performing voice recognition, using sub-word models dependent upon adjacent environments, such as a try phone. SOLUTION: When calculating likelihoods of sound models which are referenced when calculating cumulative likelihoods of the hypotheses in voice recognition, one or more sub-word models determined depending on the adjacent sub-words are referenced by each hypothesis, and the likelihoods of sound models of respective sub-word models to a target voice of recognition are calculated, and a maximum value of calculated likelihoods is defined as the likelihood of a sound model that corresponds to the hypothesis. COPYRIGHT: (C)2007,JPO&INPIT
|