| 摘要 |
<P>PROBLEM TO BE SOLVED: To provide a speaker classification device in which linguistic features contained in a speech are utilized to detect erroneous classification caused by sonic features. <P>SOLUTION: A speaker classification device of an embodiment comprises sonic classification means, linguistic feature extraction means and erroneous classification detection means. The sonic classification means classifies speeches contained in inputted sonic signals for each speaker by utilizing sonic features extracted from the sonic signals and acquires a classification result. The linguistic feature extraction means acquires a character string representing contents of the speech and utilizes the character string to extract linguistic features of the speaker contained in the classification result. The erroneous classification detection means discriminates whether or not the character string representing the contents of the speech meets the linguistic features of the speaker classifying the speech in the classification result and, if not, discriminates the speech as a speech erroneously classified by the sonic classification means. <P>COPYRIGHT: (C)2013,JPO&INPIT |
