MIT CogNet, The Brain Sciences ConnectionFrom the MIT Press, Link to Online Catalog
SPARC Communities
Subscriber : Stanford University Libraries » LOG IN

space

Powered By Google 
Advanced Search

 

Predicting Palatal Contacts from Jaw and Tongue Commands: A New Sensory Model and Its Potential Use in Speech Control

 Jean-Luc Schwartz and Louis-Jean Boë
  
 

Abstract:

A number of models begin to appear in the literature, linking perception and action in a quantitative computational framework. However, in this framework, models of tactile sensors are badly lacking, though their role is generally considered as important in speech control. The purpose of this paper is to present a preliminary model of palatal tactile perception. First, we propose a review of a number of behavioural and neurophysiological data about orosensorial inputs and their potential role in speech production, particularly in perturbed conditions. Then we introduce the principles of our model, and a precise description allowing its reimplementation by anybody interested. Our departure point is a series of palatographic data on vowels and plosives in vocalic context published in the literature (Recasens 1991; Recasens et al., 1993), associated to vowel and plosive articulatory prototypes defined on an articulatory model (Maeda, 1990; Boë et al., 1995). From these data ! we derived a model enabling to predict the main characteristics of palatal contact from the articulatory commands. We show that the model provides a good approximation of palatographic data. It allows to predict both prototypical configurations and a large set of variations around these configurations due to coarticulation and tongue-jaw compensations. Then, to demonstrate the interest of this model, we realised a large articulatory-acoustic-"palatographic" dictionary (with more than 1,000,000 sets) and we selected all configurations providing a configuration of palatal contacts around the one for our prototype of [i]. We show that the configurations are all associated with the same values of these "acoustically significant area-function information" mentioned by Gay et al. (1981), that is the location and area at the point of maximum constriction: typically, with a small front constriction. Hence this model of tactile sensor could provide a very important set of orosensory in! formation useful for speech acquisition and speech control in normal and perturbed conditions.

 
 


© 2010 The MIT Press
MIT Logo