SPEECH MODEL STORING METHOD OF THE MULTI-MODEL BASED SPEECH RECOGNITION

摘要

A speech model storing method of a speech recognizer, according to an embodiment, is to easily generate and store a compressed speech module corresponding to a sound signal expressed. The speech model storing method of the speech recognizer comprises the steps of: generating a text corpus corresponding to the sound signal expressed; generating an ARPA speech model in community standard ARPA format based on the text corpus; separating the ARPA speech model into a vocabulary speech model and a ngram speech model; and generating and storing a compressed speech model by encoding the vocabulary speech model and the ngram speech model following a set speech model score calculation algorithm.