| 摘要 |
<p>PURPOSE:To enable speech recognition with high accuracy by applying dynamic programming for the positioning of a phoneme pattern for an input speech. CONSTITUTION:A CPU 20 controls a processing procedure, data from a RAM 23, floppy disk 24, or ROM 21 for phoneme pattern storage are used to classify the segmentation of phonemes roughly, and suitable recognizing methods are used for the respective phonemes to recognize them. Further, a symbol series developed by a word dictionary 22 and the input voice are matched on a dynamic programming basis by using a table containing phoneme recognition results so as to find phoneme borders in detail. Points obtained by the dynamic programming are compared to determine a recognized word finally. Consequently, the difficulty in phoneme recognition and difficult in a case wherein many and unspecified persons are included are eliminated to enable the speech recognition with high recognition accuracy.</p> |
