摘要 |
<P>PROBLEM TO BE SOLVED: To construct an acoustic model with high accuracy. <P>SOLUTION: An sound model adaptation device includes selection means for selecting, from among plural speech recognition data pieces, test data for speech recognition and learning data for learning an acoustic model, first adaptation means for performing preliminary adaptation processing of the acoustic model using learning data, speech recognition means for performing speech recognition processing on the basis of the acoustic model subjected to adaptation processing by using test data, first determination means that compares the speech recognition results with plural speech recognition data pieces to determine whether the recognition result accuracy has improved or not, confidence level changing means for changing confidence level of speech recognition data on the basis of the determination results, data determination means that determines speech recognition data on the basis of the changed confidence level, and second adaptation means for performing final adaptation processing of the acoustic model by using the determined speech recognition data. <P>COPYRIGHT: (C)2013,JPO&INPIT |