PRONOUNCING FLUCTUATION RULE EXTRACTION DEVICE, PRONUNCIATION FLUCTUATION RULE EXTRACTION METHOD AND PRONUNCIATION FLUCTATION RULE EXTRACTION PROGRAM

摘要

A problem to be solved is to robustly detect pronunciation fluctuation cases with less labor to obtain a pronunciation fluctuation rule with high general versatility. The problem is solved by a pronunciation fluctuation rule extraction device provided with a voice data memory unit, a standard type pronunciation memory unit, a sub-word language model generating unit, a voice recognizing unit, and a differential extracting unit. The voice data memory unit stores voice data. The standard type pronunciation memory unit stores standard type pronunciation data indicative of standard pronunciation of voice data. The sub-word language model generating unit generates a sub-word language model from the standard type pronunciation data. The voice recognizing unit recognizes the voice data by using the sub-word language model. The differential extracting unit compares a recognition result output by the voice recognizing unit and the standard type pronunciation data to extract their difference.