Biomimetic vocal tract modeling: Synthesis of speech articulation

In recent years, a number of researchers have started to investigate cognitive aspects of speech production using “speaking robots.” This paper introduces AnTon–the Animatronic Tongue and vocal tract model developed at the University of Sheffield, UK. AnTon’s distinctive feature is its design principle. Whereas other speaking robots are motivated by functionality—to produce a good imitation of human speech—AnTon mimics human anatomy, including tissue properties and muscle (actuator) configurations. Any ability to produce speech sounds emerges solely as a consequence from that anatomy. Such imitation of a biological system is called “biomimetic.” The reason to choose a biomimetic approach lies in the need for an experimental tool that makes speech articulation processes transparent. The use of realistic muscle configurations means that the model’s speech gestures can be mapped directly to human behavior, facilitating the evaluation of both the quality of AnTon’s speech gesture reproductions as well as any predictions about human behavior inferred from experiments with AnTon. AnTon will be used to clarify speech articulation processes, specifically in relation to energy efficiency of articulatory gestures and the importance of an auditory feedback loop. The project’s progress is continually documented on http://www.dcs.shef.ac.uk/∼robin/anton/anton.html.