Relation of vocal tract constriction location to identification of voiced stop consonants.
Brad H. Story, Kate Bunton · The Journal of the Acoustical Society of America · 2009
Consonant identification experiments are typically based on stimuli produced with a formant synthesizer where a particular acoustic characteristic is varied along some continuum of values (e.g., onset frequency of F2). In this experiment, however, vowel-consonant-vowel (VCV) stimuli were generated with a synthesizer based on a kinematic model of the vocal tract shape [B. H. Story, JASA, 117(5), 3231–3254 (2005)] where the acoustic characteristics result from the time-varying tract configuration. Each VCV was produced by superimposing a constriction at a particular location within the vocal tract on an underlying vowel-vowel sequence. Five different vowel-vowel sequences were produced, and for each, a constriction was imposed at 20 consecutive locations extending from the lips to approximately the velum. VCVs were presented in random order, ten times each, to ten listeners, using a forced choice paradigm where only the voiced English consonants /b, d, g/ were available. Listener responses indicated that the relation of formant frequencies near the consonant, to those of the preceding or succeeding vowel can provide reasonably accurate predictions of regions within the vocal tract that correspond to specific consonants. It was also revealed that the location and length of each region is dependent on the vowel. [Research supported by NIH R01-DC04789.]