SPEECH RECOGNITION SYSTEM USING TECHNOLOGY FOR IMPLICITLY ADAPTING TO SPEAKER

摘要

PROBLEM TO BE SOLVED: To attain a speech recognition method for successfully functioning even before generation and during the generation of an SD acoustic model depending on a speaker to whom unmanaged learning is applied.;SOLUTION: In order to give a speech recognition level at least equal to pure SI acoustic models 230, 232 which do not depend on a speaker, at least one SI acoustic model 230, 232 and at least one SD acoustic model 234 which depends on the speaker are combined and used. Then, acoustic templates in at least one or more acoustic models are updated by continuously performing unmanaged learning. After that, the updated SD acoustic model 234 is combined with at least one SI acoustic model 230, 232. Thus, high speech recognition performance is obtained even while a speech recognition test is performed.;COPYRIGHT: (C)2013,JPO&INPIT