Augmented classification of Japanese visemes and hierarchical weighted discrimination for visual speech recognition

摘要: For the purpose of automatic speech recognition and animation synthesis, speaker verification so on, there have been studies on `viseme'. Viseme is a visually identifiable unit utterance or equivalent in visual domain phoneme audio domain. The classification discrimination method visemes are still important topics. This paper focuses number units procedure Japanese visemes: We extend from 6 to 9 expanse word representation by their series, then propose hierarchical weighted using multiple discriminative analysis (MDA) enhance ability. In order verify discuss availability our proposals, experiments were conducted. From these results, validity proposed methods was confirmed.