A neural network approach to the dimensionality of the perceptual vowel space

Authors

  • T.M. Nearey Dept. of Linguistics, Alberta Univ., Edmonton, Alta., Canada
  • M. Kiefte

Keywords:

hearing, neural net architecture, speech recognition, neural network approach, dimensionality, perceptual vowel space, monophthongs, 2D space, F1prime, F2 prime, perceptual properties, vowels, 3D space, listener perception, three-formant vowel continuum, neural network architecture, nonparametric representation, stimulus space, 2D representation, listener behavior, listener categorization

Abstract

The question of the dimensionality of the perceptual vowel space for monophthongs has a long history. The main question we address is: Can a two-dimensional perceptual space, corresponding roughly to F1 and F2-prime adequately represent the perceptual properties of vowels? We sketch a novel method to examine the degree to which a two- or three-dimensional space can accurately represent listeners' perception of a large three-formant vowel continuum. The key to our analysis is a neural network architecture that is capable of implementing an optimal, nonparametric, two-dimensional representation of stimulus space. Our results suggest that no two-dimensional representation adequately accounts for listeners' behavior

Additional Files

Published

2003-09-01

How to Cite

1.
Nearey T, Kiefte M. A neural network approach to the dimensionality of the perceptual vowel space. Canadian Acoustics [Internet]. 2003 Sep. 1 [cited 2024 Jun. 20];31(3):16-7. Available from: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/1528

Issue

Section

Proceedings of the Acoustics Week in Canada