Speech synthesis by mimicking articulatory movements

We describe a computational model of speech production which consists of trajectory formation, for generating articulatory movements from a phoneme specific gesture, and articulatory-to-acoustic mapping for generating speech signal from the articulatory motion. The context-dependent and independent approaches in the task-oriented trajectory formation are presented form a viewpoint how to cope with the contextual variability in the articulatory movements. The model is evaluated by comparing the computed and the original articulatory trajectories and speech acoustics. Also, we describe a recovery of the articulatory motion from speech acoustics for generating the articulatory movements by mimicking speech acoustics.