SPEAKER ADAPTIVE TRAINING METHOD IN VOICE RECOGNITION SYSTEM

摘要

PURPOSE: A speaker adaptive training method in a voice recognition system is provided to generate a speaker-dependent model with respect to a specific person only with speaker independent model and a small amount of training data for a specific speaker to realize the speaker-dependent model in a potable device using HMM. CONSTITUTION: Voice data is received from a plurality of unspecified speakers(ST11), and a speaker independent model with respect to a word to be recognized is trained using the received data to generate the speaker independent model to be used as an initial model of a potable device(ST12). The potable device receives voice data of a specific speaker at least once for each word to be recognized(ST13). The speaker independent model is corrected using the voice data of the specific speaker to generate a speaker dependent model with respect to a specific person(ST14).