摘要 |
PROBLEM TO BE SOLVED: To provide a sound model adaptation technology which does not require advanced knowledge of phonology or phonetics and man power for creating a cluster, and which does not increase a size, keeping a content of the sound model as it is. SOLUTION: A phoneme model included in the sound model before adaptation which is stored in a storing section 505A is classified into a phoneme environmental non-depending model set and a phoneme environmental depending model set (510). The phoneme environmental non-depending model set is clustered for making the predetermined number of clusters (525). A phoneme environmental depending model of the phoneme environmental depending model set is clustered to a phoneme environmental non-depending model cluster which a phoneme environmental non-depending model including the phoneme environmental depending model, belongs to, and a phoneme model cluster is created (545). A parameter of the phoneme model is updated from a data for sound model adaptation and the phoneme model cluster, and the sound model after adaptation is created (555). COPYRIGHT: (C)2007,JPO&INPIT
|