摘要 |
PURPOSE: A post processing method is provided to improve a rejection performance of a speech recognition system by the usage of a centroid representative of each state of a model. CONSTITUTION: A post processing method using a centroid in a speech recognition system comprises the steps of: determining whether a candidate word is recognized or rejected by the usage of a centroid representative of each state of a model and a view vector in a best sequence(ST7); if the candidate word is determined as a recognition rejection, abandoning a first candidate word to require a re-pronunciation to a user(ST6); and if the candidate word is determined as recognition, determining the first candidate word as a recognized word(ST7). |