| 摘要 |
PROBLEM TO BE SOLVED: To provide a voice recognition device which can maintain a voice recognizing rate to non-specified speakers while raising a voice recognizing rate to specified speakers. SOLUTION: A voice recognition part 7 calculates incidence probability with which a series of feature parameters occurs by referring to feature parameters detected in a feature value detecting part 3, a Hidden Markov Model (HMM) for non-specified speakers and a Hidden Markov Model(HMM) for specified speakers. That is, the voice recognition part 7 is made to have state series of one hidden Markov model common to the HMM for non-specified speakers and the HMM for the specified speakers as to each word in a process for calculating occurrence probability and, moreover, at the time of calculating the probability of the respective states of the one HMM, the part 7 carries out the multiplication of output stochastic function values and transition probability of two sets whose combination is preliminarily set among output stochastic function values and transition probability stored in data storage parts 4, 6, and the part 7 selects a larger one between these multiplication values, and the part 7 calculates occurrence probability based on selected multiplication values to perform the recognition of an input voice based on the calculated occurrence probability. COPYRIGHT: (C)2006,JPO&NCIPI
|
