An auditory-feedback-based neural network model of speech production that is robust to developmental changes in the size and shape of the articulatory system
De. Callan et al., An auditory-feedback-based neural network model of speech production that is robust to developmental changes in the size and shape of the articulatory system, J SPEECH L, 43(3), 2000, pp. 721-736
The purpose of this article is to demonstrate that self-produced auditory f
eedback is sufficient to train a mapping between auditory target space and
articulator space under conditions in which the structures of speech produc
tion are undergoing considerable developmental restructuring. One challenge
for competing theories that propose invariant constriction targets is that
it is unclear what teaching signal could specify constriction location and
degree so that a mapping between constriction target space and articulator
space can be learned. It is predicted that a model trained by auditory fee
dback will accomplish speech goals, in auditory target space, by continuous
ly learning to use different articulator configurations to adapt to the cha
nging acoustic properties of the vocal tract during development. The Maeda
articulatory synthesis part of the DIVA neural network model (Guenther et a
l., 1998) was modified to reflect the development of the vocal tract by usi
ng measurements taken from MR images of children. After training, the model
was able to maintain the 11 English vowel targets in auditory planning spa
ce, utilizing varying articulator configurations, despite morphological cha
nges that occur during development. The vocal-tract constriction pattern (d
erived from the vocal-tract area function) as well as the formant values va
ried during the course of development in correspondence with morphological
changes in the structures involved with speech production. Despite changes
in the acoustical properties of the vocal tract that occur during the cours
e of development, the model was able to demonstrate motor-equivalent speech
production under lip-restriction conditions. The model accomplished this i
n a self-organizing manner even though there was no prior experience with l
ip restriction during training.