Training a Vocal Tract Synthesiser to imitate speech using Distal Supervised Learning
Ian Spencer Howard, Mark A. Huckvale · UCL Discovery (University College London) · 2005
Imitation is a powerful mechanism by which both animals and people can learn useful behavior, by copying the actions of others. We adopt this approach as a means to control an articulatory speech synthesizer. The goal of our project is to build a system that can learn to mimic speech using its own vocal tract. We approach this task by training an inverse mapping between the synthesizer’s control parameters and their auditory consequences. In this paper we compare the direct estimation of this inverse model with the distal supervised learning scheme proposed by Jordan & Rumelhart (1992). Both of these approaches involve a babbling phase, which is used to learn the auditory consequences of the articulatory controls. We show that both schemes perform well on speech generated by the synthesizer itself, when no normalization is needed, but that distal learning provided slightly better performance with speech generated by a real human subject.