The Perceptual Significance of a Relative Acoustic Representation of Speech

Abstract

The acoustic signature of a particular speech sound varies according to its surrounding phonetic context. How listeners overcome these context dependent acoustic differences is not well understood. Researchers have attempted to find a one-to-one relationship between speech sounds and their acoustic realizations for over 50 years. More recently, relative acoustic representations have been proposed as a means of addressing the lack of acoustic invariance problem. Unlike absolute acoustic cues that are dependent on the context of a given speech signal, relative acoustic representations consist of abstract acoustic patterns that are context independent. The primary goal of this dissertation was to evaluate the perceptual significance of a relative acoustic representation of speech called a relative formant deflection pattern and formally test the relative formant deflection pattern hypothesis. A series of experiments were used to 1) establish how the relative formant deflection pattern hypothesis is different from other relative acoustic hypotheses of place of articulation perception and 2) explicitly test the ability of the relative formant deflection pattern hypothesis to predict listeners' place of articulation perceptions. Chapter 2 established that the relative formant deflection pattern hypothesis significantly predicted listeners' identifications more often than a locus-equation-inspired hypothesis in three out of four vowel contexts. Chapter 3 validated a novel acoustic measurement capable of extracting relative formant deflection patterns from recorded speech and then used the developed tool to evaluate listeners' perceptions of relative formant deflection patterns in natural, reduced speech. The experiment found that the relative formant deflection pattern hypothesis correctly predicted the dominant relative formant deflection pattern that solicited listeners' %/b/ identifications and %/g/ identifications, but not listeners' %/d/ identifications. Chapter 4 established that in a more controlled experimental setting the relative formant deflection pattern hypothesis accounted for listeners’ phonetic identifications across vowel contexts and speech manipulation conditions (i.e. simulated versus sine-wave speech). Additionally, the final experiment in Chapter 4 established that relative deflection in all three formant frequencies was necessary to fully account for the predictive power of the relative formant deflection pattern hypothesis. Taken together, the research provides insight into the perceptual significance of different relative acoustic representations of speech and proposes that a relative acoustic representation called a relative formant deflection pattern may be a perceptually significant solution to the lack of acoustic invariance problem.