Synthesis of french fricatives by audio-video to articulatory inversion

摘要

This paper presents an articulatory approach to the synthesis of fricative consonants in vocalic context. The articulatory, aerodynamic and acoustic components of a complete articulatory synthesiser and their interactions are briefly described. The articulatory trajectories of the control parameters for the synthesiser are estimated by inversion from audio-video recordings of the reference subject based on whom the synthesiser was elaborated. The inversion method is described and evaluated. For synthesis, a simple strategy of control of the glottis gesture, in coordination with the oral constriction, was used to produce both voiced and voiceless fricatives. The quality of the acoustic synthesis has been assessed by a forced choice identification test which led to very similar scores for synthetic and natural stimuli. These results validate at the same time the realism of the articulatory synthesiser and the strategies used to control it. Articulately synthesis appears thus to be able to produce high quality speech sounds, and to provide the ground for the development of a virtual talking head. Moreover, the articulatory data obtained by inversion and the methodology developed will serve as the basis for studying human control strategies for speech production. [References: 41]