VOICE RECOGNITION SYSTEM USING IMPLICIT SPEAKER ADAPTION

摘要

PROBLEM TO BE SOLVED: To provide a voice recognition method that excellently functions even before and during generation of a speaker dependent (SD) acoustic model to which unsupervised training is applied. SOLUTION: A voice recognition (VR) system is disclosed that utilizes at least one speaker independent (SI) acoustic model 230 or 232 in combination with at least one speaker dependent (SD) acoustic model 234 to provide a level of speech recognition performance that at least equals that of the purely SI acoustic model 230 or 232. The disclosed VR system continually uses unsupervised training to update the acoustic templates in the one or more acoustic models. The VR system then uses the updated SD acoustic model 234 in combination with the at least one SI acoustic model 230 or 232. Consequently, improved VR performance is obtained even during VR testing. COPYRIGHT: (C)2008,JPO&INPIT