| 摘要 |
PURPOSE:To provide for high speed retrieval by storing histories of dictionary retrieval and performing dictionary retrieval only as to the subsequent portion. CONSTITUTION:In a phoneme identifying part 1, phoneme identification is performed of the input voice to output the results of identification and the values of probabilities. A sentence candidate producing part 2 produces plural sentence candidates from all the combinations of the obtained phoneme matrices from the least value of the sum total of the probabilities. A sentence candidate classification part 3 re-arranges these in a predetermined sequence. A comparator part 4 reads the content of a grammar dictionary 6 indicated by a dictionary pointer 9 and performs magnitude comparison with respect to all the sentence candidates and outputs the result of coincidence to a result adjudication part 10. In case of non-coincidence, the relative magnitude is written in a retrieval flag memory part 5 while the results of comparison are outputted to a counter control part 7. The control part 7 issues count down or count up instructions based on the results. A flag counter 8 shifts the retrival flag memory position pointer by one by the instructions of the control part 7. |
