ERROR CORRECTION FOR SPEECH RECOGNITION SYSTEMS Russian patent published in 2010 - IPC G10L15/00 

FIELD: physics; computer engineering.

SUBSTANCE: invention relates to application program products, designed for correcting words in a word sequence which is obtained during recognition of an input speech sequence. Words in a word sequence obtained after recognition of an input speech sequence are presented to a user, and one of the words in the sequence is replaced if selected for correction by the user. Words with low value of recognition reliability are selected; alternative candidate words for one of the selected words are ordered in accordance with an ordering criterion; after replacing words, the order of alternative candidate words for neighbouring words in the sequence can be updated; the word which is replaced can be obtained from the speech presentation of one of the selected words through speech recognition with a limited vocabulary; and the word which replaces one selected word can be obtained from the speech and literal presentation of one selected word.

EFFECT: increased mean accuracy of connected-speech recognition.

28 cl, 13 dwg