STANDARD PATTERN GENERATING METHOD FOR SPEECH RECOGNITION, RECORDING MEDIUM FOR CONTROL PROGRAM FOR STANDARD PATTERN GENERATION FOR IMPLEMENTING THE SAME METHOD, AND STANDARD PATTERN GENERATING DEVICE USING THE SAME METHOD

摘要

PROBLEM TO BE SOLVED: To improve the voice recognition precision. ;SOLUTION: When voice data spoken more than once by speakers as to an arbitrary word are inputted, a feature pattern extraction part 1 divides each voice data into frames having equal time intervals, extracts feature vectors of voice waveforms in each frame, and outputs time-series data of respective feature vectors as a feature pattern. A phoneme position adjustment part 2 selects an arbitrary feature pattern as a standard pattern and corrects other feature patterns so that feature vectors of the respective patterns correspond to the standard vector of the standard pattern. The corrected feature patterns are supplied to a standard pattern generation part 3, which generates the standard pattern free of influence of variance in the border positions of phonemes between voice data.;COPYRIGHT: (C)1999,JPO