A rough guide to the acoustic-to-articulatory inversion of speech

Abstract | This article reviews a specic speech research area called acoustic-to-articulatory inversion of speech, or speech inversion, which refers to the problem of mapping the acoustic speech signal onto a space describing the con-guration of the human vocal tract that actually produced this signal. This space may be modeled in a variety of ways, such as with trajectories of the movement of the articula-tors- certain parts of the human vocal tract- derived by means of some specialized medical imaging device, or by us-ing linguistics-related abstract classes to describe the evolu-tion of the articulatory state through time. The problem is by far non-trivial, mainly due to the one-to-many nature and the high non-linearity of the acoustic to articulatory map-ping. Numerous approaches towards its solution have been proposed. Such a solution could have several applications, with most important the probability of improving the per-formance of current automatic speech recognition systems