Does physical realism of articulatory modelin improve the perception of synthetic speech?

Daniel Pape (1, 2) , Pascal Perrier (3) , Susanne Fuchs (4) , Sonia Kandel (5)

1 IEETA - Institute of Electronics and Telematics Engineering of Aveiro
2 GIPSA-lab - Grenoble Images Parole Signal Automatique
3 GIPSA-MAGIC - GIPSA - Machines parlantes, Gestes oro-faciaux, Interaction Face-à-face, Communication augmentée
4 ZAS - Zentrum für Allgemeine Sprachwissenschaft [Berlin]
5 LPNC - Laboratoire de Psychologie et NeuroCognition

This article presents the first step in a process to evaluate the potential impact of the physical properties of the articulators on speech perception. The underlying hypothesis is that articulatory biomechanics contribute to shape articulatory and formant trajectories. These in turn become patterns available for speech perception. Perceptual tests on synthetic silent center stimuli were run which are inspired from former studies of Strange and colleagues. Stimuli were generated with a unique timing with different models incorporating various degrees of physical realism. Our results show that silent center stimuli generated with a realistic biomechanical model achieve higher identification scores than stimuli from less realistic, kinematic, models, when only fast reaction times of the listeners are considered, in order to make sure that only low-level cognitive processings are involved.

Keywords: Speech production; muscle models; biomechanical models

Domains: Neuroscience; Cognitive Sciences; Linguistics; Linguistics

Daniel Pape, Pascal Perrier, Susanne Fuchs, Sonia Kandel. Does physical realism of articulatory modelin improve the perception of synthetic speech?. ISSP 2011 - 9th International Seminar on Speech Production, Jun 2011, Montréal, Canada. pp.153-154. ⟨hal-00605683⟩