VOICE RECOGNITION DEVICE, VOICE RECOGNITION METHOD, VOICE RECOGNITION PROGRAM, AND RECORDING MEDIUM

摘要

In a sound recognition apparatus (500), upon receiving a digital sound signal, an obtaining unit (501) obtains acoustic feature quantities A and B concerning an amount of spectrum information and an amount of attribute information. Next, an element-distribution-likelihood calculating unit (502) calculates an element distribution likelihood of each subword for each of attributes. An identifying unit (506) identifies a category of an unknown sound, and a weighting unit (507) refers to a lookup table (700) for the identified category to weight a branch probability of each subword for each of the attributes. A subword-likelihood calculating unit (504) calculates a subword likelihood using the element distribution likelihood of each subword for each attribute, and the branch probability of each subword after the weighting process for each attribute. Finally, a word-likelihood calculating unit (511) calculates a word likelihood and outputs a word having the highest word likelihood as a result of recognition.