Simulating Vocal Imitation in Infants, using a Growth Articulatory Model and Speech Robotics

摘要

In order to shed lights on the cognitive representations\nlikely to underlie early vocal imitation, we tried to simulate\nKuhl and Meltzoff's experiment (1996), using Bayesian\nrobotics and a statistical model of the vocal tract that had\nbeen fitted to pre-babblers' actual vocalizations. It was\nshown that audition is compulsory to account for infants'\nearly vocal imitation performance, inasmuch as the\nsimulation of purely visual imitation failed to reproduce\ninfants' score and pattern of imitation. Further, a small\nnumber of vocalizations (less than 100!) appeared to be\nenough for a learning process to provide scores at least as\nhigh as those of pre-babblers. Thus, early vocal imitation\nlies in the reach of a baby robot, with only a few\nassumptions about learning and imitation.