To provide a voice recognition device capable of performing acoustic recognition for voice continuously generated and of obtaining the recognition results discriminating different expressed words.
An acoustic recognition unit 11 creates information of plural candidates regarding an array of acoustic units, a syntax analysis unit 13 judges whether or not the array of each acoustic unit follows a syntax rule for information of plural candidates and creates the array of plural vocabulary items corresponding to the array of each acoustic unit. After a different expressed word is expanded from the array of the word items by means of an different expressed work expansion unit 20, each information evaluation value generation unit 16 determines an evaluation value of that probability corresponding to the array of plural vocabulary items. An integral evaluation value determination unit 17 weights, adds, and integrates each information evaluation value and determines each integral evaluation values corresponding to the array of plural vocabulary items. A recognition result creation unit 18 arranges a candidate of the array of plural vocabulary items in order of each integral evaluation value determined and outputs the recognition results of input voice.
SUMIYA KAZUHIKO
UIRIAMU GANON
HAYAKAWA AKIHIKO
KUNO SUSUMU
TOOMASU RINCHI
MIZUNO JUNKO
SAMIYUERU PIRATO
JIYON DOURII
KURZWEIL APPLIED INTELLIGENCE
Next Patent: ADAPTIVE CONVERSION ENCODING AUDIO SIGNAL DECODER