SPEECH RECOGNITION SYSTEM

摘要

PROBLEM TO BE SOLVED: To provide a speech recognition system in which the recognition rate of speech recognition in noisy environment is improved by performing adaptive recognition. SOLUTION: A feature vector extraction part 30 extracts feature information according to speech corresponding to utterance. A search part 60 receives the output of the feature extraction part and makes a search wile matching it against previously set sound models to perform speech recognition. The search part 60 uses a Monte Carlo particle filter to perform the estimation of noise parameters and the time development of a hidden state constituting a hidden Markov model, frame by frame, in parallel.