ISCA Archive Interspeech 2023
ISCA Archive Interspeech 2023

Comparing /b/ and /d/ with a Single Physical Model of the Human Vocal Tract to Visualize Droplets Produced while Speaking

Takayuki Arai, Tsukasa Yoshinaga, Akiyoshi Iida

The BMW-RL model, a physical model of the human vocal tract that produces /b/, /m/, /w/, /r/, and /l/, has been utilized to investigate not only each single sound but also consonant clusters such as /br/. This model started gaining attention in 2021 in light of the COVID-19 pandemic, as it can demonstrate how droplets are expelled from the lips when producing the /b/ sound by applying a laser sheet. In this study, we redesigned the model to produce both /b/ and /d/, since both are voiced plosives and the only difference is the place of articulation. With the original BMW-RL model, the first half of the tongue rotates and produces /r/ and /l/, while in the newly proposed model, the width of the tongue is wide enough to make a complete closure at the alveolar position for producing /d/. We tested how the different place of articulation affects the ways of expelling droplets by using the single model producing /b/ and /d/ and found that more droplets were expelled with /b/ than /d/.