VOICE RECOGNITION DEVICE AND VOICE RECOGNITION PROGRAM

摘要

PROBLEM TO BE SOLVED: To provide a voice recognition device which realizes a real-time recognition processing within the frame length by reducing the processing volume of an output probability calculation part in a large-vocabulary voice recognition system. SOLUTION: The relation between phonemes of a current frame and phonemes expected to occur in the next frame is stored in a consecutive phoneme information storage part 14. A phoneme selection part 15 selects a phoneme of high output probability in the current frame and extracts a phoneme expected to occur in the frame following the selected phoneme from a phoneme information storage part 11 on the basis of consecutive phoneme information in the consecutive phoneme information storage part 14 and stores the selected phoneme in a selected phoneme information storage part 16. Only the phoneme stored in the selected phoneme information storage part 16 is subjected to output probability calculation in the next frame to reduce the processing volume in an output probability calculation part 10.