Target speaker learning method , the device and program

摘要

PROBLEM TO BE SOLVED: To suppress natural deterioration of synthetic speech of a target speaker.;SOLUTION: A plurality of clusters are set for every feature by independently clustering feature amounts of previously obtained speech data of a plurality of speakers for every feature. A combination of clusters to which a combination of feature amounts of the speech data of a target speaker belongs is selected and speech data corresponding thereto are selected from the speech data of the plurality of speakers. When the combination of clusters to which the combination of feature amounts of the speech data of the target speaker belongs and the combination of the feature amounts of the selected speech data are different from the combination of the cluster, partial feature amounts of the feature amounts of the selected speech data are converted.;COPYRIGHT: (C)2013,JPO&INPIT