摘要 |
PROBLEM TO BE SOLVED: To provide an adaptive model generating method for speech recognition which lightens the load on a user who generates adapted models adapted to respective users and use environments. SOLUTION: Provided are a sufficient statistics and selection model generating means which generates sufficient statistics and selection models for selecting the sufficient statistics by using 1st speech data, a sufficient statistics and selection model storage means, a plurality of data generating means which generate a plurality of pieces of (K)th feature value data by using 2nd speech data as voice data of a user, a (K)th adaptive model generating means which selects sufficient statistics acoustically close to the (K)th feature value data by using a selected model among the sufficient statistics and generates (K)th adaptive model candidates by using the selected sufficient statistics, an adaptive model candidate storage means which stores the (K)th adaptive model candidates, and an adaptive model selecting means which selects an adaptive model candidate acoustically close to 2nd speech data as voice data of the user. COPYRIGHT: (C)2003,JPO
|