SPEECH RECOGNITION DEVICE, SPEECH RECOGNITION METHOD AND RECORDING MEDIUM

摘要

PROBLEM TO BE SOLVED: To prevent speech recognition precision from being degraded due to unknown words. SOLUTION: A dictionary database 6 stores the words being objects of speech recognition and also a word dictionary in which suffix words being phonemes and phoneme trains constituting unknown words and classify the unknown words for every part of speech are registered. A matching section 4 connects acoustic models of an acoustic model database 5 based on the word dictionary and computes a score using a group of the featured values outputted by a feature extracting section 3 based on the connected acoustic models. Then, the section 4 selects a series of words being speech recognition results based on the score.