VOICE PROCESSING DEVICE, VOICE PROCESSING METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To easily identify a speaker without previously registering speakers.SOLUTION: A feature quantity extraction unit 220 extracts the feature quantity in the vowel section of voice data. A feature quantity classification unit 230 classifies the extracted feature quantity into clusters as many as the speakers for each vowel by a no-teacher learning method. A combination determination unit 240 determines a combination corresponding to the same speaker, of the combinations obtained by extracting each cluster for each vowel. A partition unit 250 partitions the voice data into respective speakers in accordance with the appearance position of the vowel section of the feature quantity included in the combination corresponding to the same speaker.