| 摘要 |
PROBLEM TO BE SOLVED: To provide a voice recognition device in which a word having been newly registered in a language model is recognized with high accuracy.SOLUTION: A voice recognition device 10 includes: a stochastic language model storage part for storing a stochastic language model which holds a set of words and a parameter attendant on each of the words; and a language model updating part 3 for determining a parameter attendant on a new registration word by using a new registration word, which is a word newly registered, and voice data including the new registration word, and registering the new registration word and the parameter attendant on the new registration word in the stochastic language model; wherein the language model updating part 3 identifies a parameter by which a new registration word part in the voice data is correctly recognized when determining a parameter attendant on a new registration word, and determines an identified parameter as a parameter attendant on the new registration word. |
