SPEECH RECOGNITION METHOD USING HIDDEN TRAJECTORY HIDDEN MARKOV MODEL

摘要

PURPOSE: A speech recognition method using hidden trajectory hidden Markov model is provided to produce a speech recognition framework capable of carrying out rapid decoding in consideration of a variation in feature trajectory caused by a speech producing mechanism because of various speaking styles. CONSTITUTION: A speech recognition method determines vocal tract response frequencies for the state in a specific frame based on vocal tract response frequencies determined for the previous frame. The vocal tract response frequencies are used to decide probability distribution for the state. Probability about observation values received for the frame is decided from the probability distribution. The vocal tract response frequencies are determined using noise-free recursive definition. The use of the recursive definition improves a decoding speed.