摘要 |
PROBLEM TO BE SOLVED: To provide a voice recognition device capable of recognizing a voice at high speed by using an inexpensive memory with less capacity capable of being operated at a high speed. SOLUTION: An input pattern extracted from an input voice by an analysis part 1 is stored in an input pattern memory 2, and the distance between a sub-word standard pattern stored in a sub-word standard pattern memory 4 and the input pattern is measured by a distance calculation part 3 to be stored in a first distance memory 5. A distance extraction part 6 reads out a required distance value from the first distance memory 5 based on sub-word line information of a recognition subject word in a dictionary memory 9 to block transfer it to a second distance memory 8. A recurrence formula calculation part 7 performs recurrence formula calculation in word by using the distance value in the second distance memory 8 and a cumulative value in a cumulative memory 10, and a result decision part 14 recognizes the word from storaged contents in the cumulative value memory 10 to output it. |