Modeling the acoustics of American English /r/ using configurable articulatory synthesis (CASY)

The claim that articulatory variation in /r/ production exhibits systematic tradeoffs to achieve a stable acoustic signal (Guenther et al., 1999) was tested using configurable articulatory synthesis (CASY) and ultrasound data. In particular, the hypothesis was tested that multiple constrictions during /r/ production are necessary to achieve a low enough F3. Ultrasound and Optotrak data from four speakers pronouncing /r/ in different vocalic contexts were used to determine where in the vocal tract the tongue gestures are placed. This data was then modeled using CASY parameters and was used to determine how the three gestures in /r/ (labial, palatal, and pharyngeal) contribute to the F3 value observed in the speech signal simultaneously recorded with the ultrasound. This was done by varying the degree and location of the lingual constrictions and the degree of the labial constriction and determining the effect on F3. It was determined that the three gestures in /r/ contribute in differing amounts to the overall F3 lowering. Furthermore, it does not seem that all three gestures are necessary for F3 lowering. This lends support to the hypothesis that the goal in /r/ production is the simultaneous achievement of three gestures. [Work supported by NIH Grant DC-02717.]