Abstract

This study examines emotion-specific information (ESI) in the articulatory movements estimated using acoustic-to-articulatory inversion on emotional speech. We study two main aspects: (1) the degree of similarity between the pair of estimated and original articulatory trajectories for the same and different emotions and (2) the amount of ESI present in the estimated trajectory. They are evaluated using mean squared error between the articulatory pair and by automated emotion classification. This study uses parallel acoustic and articulatory data in 5 elicited emotions spoken by 3 native American English speakers. We also test emotion classification performance using articulatory trajectories estimated from different acoustic feature sets and they turn out subject-dependent. Experimental results suggest that the ESI in the estimated trajectory, although smaller than that in the direct articulatory measurements, is found to …

Date

2012

Authors

Jangwon Kim, Prasanta Ghosh, Sungbok Lee, Shrikanth S Narayanan

Conference

Proceedings of The 2012 Asia Pacific Signal and Information Processing Association Annual Summit and Conference

Pages

1-4

Publisher

IEEE