Discrimination between natural and unnatural articulations based on articulatory structure

One-to-many relationship between speech sound and articulation brings an ill-posed problem in inverse estimation of articulation from speech sound. Due to the one-to-many problem, some unnatural articulations are included among the candidates in the results of the inverse estimation. To exclude the unnatural articulations for solving the one-to-many problem in inverse estimation, it is necessary to discriminate natural and unnatural articulations. For this purpose, this paper proposes a discrimination method based on articulatory structure, and discriminates between natural and unnatural articulations using the proposed method. As a result, unnatural articulations could be discriminated from natural articulations with accuracy more than 99%. This suggests the possibility that the estimation accuracy of inverse estimation can be improved by applying the proposed method.