Konferans bildirisi Açık Erişim

Database construction for speech to lip-readable animation conversion

2006    acs, Gyorgy Ta; Tihanyi, Atilla; Bardi, Tamas; Feldhoffer, Gergo; Srancsi, Balint

The training database was one of the critical element in our speech to facial animation conversion system. This system was developed as a communication aid for deaf people. The specific database was constructed from audio and visual records of professional lip-speakers. The standardized MPE -4 system was used to animate the talking head model. The trained neural net is able to calculate with acceptable error the principal component weights of feature points from the speech frames. The feature point coordinates are, calculated from PC weights. The whole system can be implemented in mobile phones. Deaf persons were able to recognize about 50 of words from the speech driven animation in the final test.