| 摘要 |
PURPOSE:To provide a voice recognition device capable of dealing with the generation of a meaningless word and natural utterance to generate plural sentences continuously by performing syntax analysis based on the constraint of syntax setting each word candidate outputted from a voice recognizing means in time series as a key word. CONSTITUTION:A voice feature parameter system generated by a voice input part 1 is sent to a voice recognition part 2. The voice recognition part 2 is coupled with voice recognition dictionary memory 21 which stores the voice feature pattern of the key word that is a recognition target word decided in advance, and collates the dictionary of the voice recognition dictionary memory 21 with the voice feature parameter system. The recognition result of the voice recognition part 2 is sent to a syntax and meaning analysis part 3 in the form of a word system (word lattice) discrete in point of time. The syntax and meaning analysis part 3 inputs the key word at every detection of the key word from the voice recognition part 2, and performs syntax and meaning analysis by using grammar and dictionary information in which the syntax meaningful information and temporal relation of an input text stored in a grammar and dictionary storage part 4 are described. |
