4pSCa8. Modeling the acoustics of American English /r/.

Session: Thursday Afternoon, June 19

Author: Carol Y. Espy-Wilson
Location: Dept. of Elec. and Comput. Eng., Boston Univ., Boston, MA 02215
Author: Narayanan Shrikanth
Location: Univ. of California, Los Angeles, CA, 90024
Author: Abeer Alwan
Location: Univ. of California, Los Angeles, CA, 90024
Author: Suzanne E. Boyce
Location: Boston Univ., Boston, MA 02215


It is often assumed that speakers use idiosyncratic combinations of constrictions at the lips, pharynx, and along the hard palate to achieve the low F3's---typically between 1300 and 1800 Hz---characteristic of American English /r/. This account was tested in a modeling study using attested vocal tract dimensions derived from three-dimensional magnetic resonance imaging data from four speakers [Alwan et al., J. Acoust. Soc. Am. (to be published) (1997)]. These dimensions were used as input to the Maeda computer model. It was determined that (1) a constriction in the post-alveolar region of the palate and at least one other constriction are needed to bring F3 into the 1300- to 1800-Hz range, (2) the maximum effect of the lip constriction was around 100 Hz, while that of the pharyngeal constriction was around 200 Hz, and (3) the effects of changing constrictions were not additive. Over all manipulations, the lowest F3 achieved was 1495 Hz, which is high compared to many /r/ productions. Using ``retro- flex'' dimensions, the lowest F3 achieved was 2224 Hz. It may be that additional acoustic mechanisms are required to account for the full range of acoustic profiles seen for /r/. [Work supported by NSF and NIH.]

ASA 133rd meeting - Penn State, June 1997