摘要 |
<p><P>PROBLEM TO BE SOLVED: To additionally enhance authentication performance in a voice authentication system which performs authentication based on voice. <P>SOLUTION: By using a feature parameter from a first acoustic analysis section 102, the learning of an HMM acoustic model is performed according to a prescribed algorithm in a speaker HMM acoustic model creation section 114 for normalization and mapping is performed in a two-dimensional acoustic space by a two-dimensional space mapping section 115 and is zone-divided by a zone division section 116. As to the respective zones, the feature parameters of the speakers for normalization belonging to the respective zones are read out of a memory section 113 for the feature parameters of the speakers for normalization by a zone HMM acoustic model creation section 117 and the creation of the HMM acoustic model is performed according to the the prescribed algorithm and is memorized into a memory section 104 for the HMM acoustic model for normalization. <P>COPYRIGHT: (C)2006,JPO&NCIPI</p> |