Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Vowel Quality Assessment Based on Analysis of Distinctive Features

Shuping Ran, Bruce Millar, Iain Macleod

Computer Sciences Laboratory, Research School of Information Sciences and Engineering, Australian National University, Canberra, Australia

This paper presents a novel method for vowel quality assessment which is based on the analysis of distinctive features. A sub-set of the features proposed by Jakobson, Fant and Halle are used. The acoustic features associated with each selected feature are encoded in the weights of a multilayer perceptron, The processing of an individual vowel sound by such feature-detecting perceptrons yields an estimate of that vowel's position in a distinctive feature space. The relative positions of one speaker's monophthongal vowel set within both a two-dimensional and a three-dimensional distinctive feature space are compared with vowel spaces based on articulatory considerations. The extent to which articulatory descriptions of acoustic vowels can be estimated using this method is discussed. Keyword: Distinctive features; Artificial neural networks, Articulatory features, Vowel quality assessment.

Full Paper

Bibliographic reference.  Ran, Shuping / Millar, Bruce / Macleod, Iain (1994): "Vowel quality assessment based on analysis of distinctive features", In ICSLP-1994, 399-402.