The gradient influence of temporal extent of coarticulation on vowel and speaker perception

Georgia Zellou,Anne Pycha

doi:10.5334/labphon.118

Georgia Zellou, Anne Pycha

Open Access

https://doi.org/10.5334/labphon.118

Copy DOI

Abstract

Coarticulation makes vowels in context acoustically different from context-free vowels. Listeners sometimes compensate by ascribing these acoustic effects to their source, but the conditions under which they do so have not yet been fully pinpointed. Ohala (1993) had suggested that acoustic effects which are temporally more distant from their source should be more susceptible to misattribution. In three experiments, we tested this hypothesis by varying the temporal extent of coda-triggered coarticulation on vowels and investigating its influence on two different perceptual behaviors: speaker-model representation and vowel-phoneme identification. Experiment 1 asked listeners to estimate speaker height based on /giC/ and /gɪC/ nonsense tokens produced by twelve female speakers. Results indicated a gradient effect: Within lax /ɪ/, greater temporal extent of coarticulation correlated with taller height judgments. Experiment 2a was similar, except that temporal extent of coarticulation in the tokens varied across a wider range of values than in Experiment 1. Results again indicated a gradient effect: Within lax /ɪ/, greater temporal extent of coarticulation correlated with taller height judgments. In Experiment 2b, listeners performed an AXB vowel-phoneme discrimination task. Results showed that greater temporal extent of coarticulation correlated with greater likelihood of listeners judging an intended /ɪ/ token to contain the vowel /ʌ/. Taken together, our results indicate that temporal extent of coarticulation affects both speaker-models and interpretation of vowel identity.

Highlights

The speech signal contains a great deal of variability, which is a potential source of ambiguity for the listener
We aim to investigate these possibilities in Experiment 2a by extending the range of variation in the extent of coarticulation in both tense /i/ and lax /ɪ/, such that their
Transition Ratio and Vowel were included to test whether temporal extent of coarticulation predicts estimated speaker height differently for lax /ɪ/ vs. tense /i/ vowels

Summary

Introduction

The speech signal contains a great deal of variability, which is a potential source of ambiguity for the listener. For the same vowel-phoneme, formants can vary by as much as 70% depending upon the context in which the vowel occurs, and by as much as 99% depending upon who the speaker is (Nearey, 1989). These large effects of context and speaker mean that, in order to successfully identify the intended vowel, the listener must do at least two things: perceptually compensate for contextual effects (Mann, 1980) and construct a model of the speaker (for an overview, see Johnson, 2008).

Objectives

Methods

Results

Discussion

Conclusion