Inherent VowelStructures in Speech Production and Perception Spaces

To examine relations between speech production and perception in the human brain, we investigated inherent vowel structures in both articulatory space and auditory perceptual space. A nonlinear analysis method (Laplacian eigenmap) was employed to extract inherent vowel structures from an articulatory database of read speech. In articulatory space, the first dimension of the vowel structure represents the degree of tongue-palate approximation, and the second is related to the ratio of lip opening to oral cavity size, while the vowels scatter on a curved surface in the third dimension. When applying the same technique to corresponding acoustic data from the same corpus, a compatible structure was obtained from the vowels in the auditory perceptual space. The results suggest that consistent topological images of the vowel structures exist in both the articulatory and auditory perceptual spaces. This finding supports the hypothesis that an efficient auditory-articulatory linkage exists in the human brain for speech processing