Mapping between acoustic and articulatory gestures

We propose a method for Acoustic-to-Articulatory Inversion based on acoustic and articulatory ‘gestures’. A definition for these gestures along with a method to segment the measured articulatory trajectories and the acoustic waveform into gestures is suggested. The gestures are parameterized by 2D DCT and 2D-cepstral coefficients respectively. The Acoustic-to-Articulatory Inversion is performed using a GMM-based regression and the results are at par with state-of-the-art frame-based methods with dynamical constraints (with an average error of 1.45-1.55 mm for the two speakers in the database).