PHONEME CLASSIFYING METHOD IN VOICE RECOGNITION SYSTEM

摘要

PURPOSE: To automate the classifying operation and improve the recognition rate of the system. ;CONSTITUTION: At S1, learning data of only one data are generated on the basis of an initial label and at S2, the phoneme recognition system is learnt from its initial state with the learning data. The learnt voice data are inputted to this phoneme recognition system to obtain a phoneme sequence at S3. At S4, the recognition rate RA is obtained from the used label and the phoneme sequence obtained at S3. At S5, DP matching between an array of phonemes of a word and the obtained phoneme sequence is performed to perform area classification, and the score is denoted as DS. At S6, a new label is obtained from the result of the area classification. At S7, learning data are generated from the obtained label and At S8, the process ends when RA=100% and DS=0, but when not, the process is repeated from S2. When the end conditions are not met within a prescribed frequency, an error log is left at S9 and corrections are made, and when the conditions are met, the obtained label is registered at S10.;COPYRIGHT: (C)1995,JPO