METHOD FOR RECOGNITION OF WORDS IN CONTINUOUS SPEECH AND DEVICE WHICH IMPLEMENTS SAID METHOD

摘要

FIELD: automation and computer engineering, in particular, voice recognition, systems for control of technological equipment, robots, computers, automatic interpretation systems, information services. SUBSTANCE: method involves elimination of phonetic decoding level and generation of hypotheses about possible words in sentence from acoustic signal directly by means of lexical decoding network which is designed as integral knowledge base which has all expected acoustic representations of given vocabulary, including phonetic transcription, phonological rules and lexicon. Corresponding device has acoustic analyzer 1, lexical analyzer 2, frequency spectral analyzer 3, spectral values storage unit 4, η weight calculation unit, current acoustic state calculation unit 6, unit 7 which traverses nodes in lexical decoding network, unit 8 which stores database of reference acoustic states, unit 9 which compares current value of functional to reference value, comparison estimations memory unit 10, control unit 11, optical estimation selector 12, edge node testing unit 13, previous nodes buffer memory unit 14, pause detector 15, unit 16 which tests sequence of nodes, memory unit 17 for recognized sequences of nodes, output unit 18. EFFECT: increased precision, increased speed of continuous voice recognition for Russian language. 2 cl, 13 dwg