APPARATUS AND METHOD FOR UTTERANCE VERIFICATION BASED ON WORD DURATION

摘要

According to the present invention, an apparatus for utterance verification using a real-time word duration modeling in a mass voice recognition system, includes: a training signal managing unit for calculating an average of phonemic duration time, and an average and a variance of word duration time of a received training signal, and modeling a relation between the average and the variance of the word duration time by using a regression analysis to generate a regression model; a voice signal processing unit for applying a received voice signal to a context-dependent phonemic model to output a voice recognition result; and an utterance verification unit for generating the average and the variance of the word duration time, or the average of the phonemic duration time based on a preset threshold of word frequency, and generating a reliability measured value based on the regression model.