| 摘要 |
PROBLEM TO BE SOLVED: To provide a voice recognition device which effectively adjust an element distribution number at a high speed as to a probability model using a mixed distribution. SOLUTION: This voice recognition device uses the probability model using the mixed distribution and composed of a standard pattern storage means 103 which holds a standard pattern, a recognition means 104 which inputs a voice and outputs the recognition result by using the standard pattern, a standard pattern generating means 102 which inputs a voice for learning and generates the standard pattern, and a standard pattern adjusting means 203 which adjusts the element distribution number of the mixed distribution of the standard pattern. Consequently, tree structures of element distributions are generated by states in voice recognition using a hidden Markov model having the mixed Gaussian distribution as an output probability distribution and the element distribution number of the respective states are adjusted by using an information amount reference.
|
