摘要 |
PROBLEM TO BE SOLVED: To suppress the deterioration of a recognition performance arising as the time after the start of a voice recognition until the start of speech becomes long by correcting a sound model by using information on environmental noises. SOLUTION: A silent sound model correction part 7 performs the adaptation of a silent sound model which is a sound model representing a silent state by a maximum likelihood method, complex statistics, or a minimal length- maximum separation theorem using input data observed in a section just before a voice recognition section as subject to the voice recognition. |