A speech recognition method using context-dependent phone duration model

摘要

PURPOSE: A speech recognition method using context-dependent phoneme duration information is provided to reduce a deterioration in a recognition rate and an increase in recognition time. CONSTITUTION: Duration information is obtained in unit of state in a context-dependent phoneme through a decision tree-based phoneme modeling training procedure. A candidate word is read to obtain context-dependent phonemes constructing the word(401). States corresponding to the context-dependent phonemes are attained on the basis of the decision tree based phoneme model, to obtain phoneme duration information using the state duration information(402,403). The word and a sentence are recognized using the phoneme duration information(405).