A computational model of perceptuo-motor processing in speech perception: learning to imitate and categorize synthetic CV syllables

Raphaël Laurent (1, 2, 3) , Jean-Luc Schwartz (2) , Pierre Bessière (1, 4) , Julien Diard (3)

1 E-MOTION - Geometry and Probability for Motion and Action
2 GIPSA-PCMD - GIPSA - Perception, Contrôle, Multimodalité et Dynamiques de la parole
3 LPNC - Laboratoire de Psychologie et NeuroCognition
4 LPPA - Laboratoire de Physiologie de la Perception et de l'Action

This paper presents COSMO, a Bayesian computational model, which is expressive enough to carry out syllable production, perception and imitation tasks using motor, auditory or perceptuo-motor information. An imitation algorithm enables to learn the articulatory-to-acoustic mapping and the link between syllables and correspond- ing articulatory gestures, from acoustic inputs only: syn- thetic CV syllables generated with a human vocal tract model. We compare purely auditory, purely motor and perceptuo-motor syllable categorization under various noise levels.

Mots clés: sensorimotor fusion; Speech perception; Bayesian modeling; computational model

Domaines: Modélisation et simulation; Informatique; Psychologie; Linguistique

Raphaël Laurent, Jean-Luc Schwartz, Pierre Bessière, Julien Diard. A computational model of perceptuo-motor processing in speech perception: learning to imitate and categorize synthetic CV syllables. Interspeech 2013 - 14th Annual Conference of the International Speech Communication Association, Aug 2013, Lyon, France. pp.2797-2801. ⟨hal-00827885⟩