Richard S. McGowan
Philip E. Rubin
Haskins Labs., 270 Crown St., New Haven, CT 06511-6695
A method involving task dynamics and a genetic algorithm [see McGowan, Speech Commun. 14, 19--48 (1994)] was used to recover articulation from the speech acoustics of a human subject. Six utterances from a single subject were chosen to test the applicability of the proposed method to human beings: /ga/, /gi/, /da/, /di/, /ba/, and /bi/. The first three formant trajectories of the natural utterances were extracted and used to represent the acoustic data. A genetic algorithm was used in constrained optimization of task-dynamic parameters applied to the Haskins Laboratories articulatory synthesizer, ASY. The articulatory movements recovered in this manner were assessed using perceptual tests of the resulting speech generated by ASY and by a visualization of the resulting ASY articulation.