Quantifying the visual salience of Swedish vowel contrasts: a computer vision approach

Description

Helene Springer, Henrik Garde, Frida Splendido, and Marianne Gullberg measure how Swedish vowels look on a speaker's lips. Seeing a face helps listeners catch speech, yet charts of lip shapes have long rested on simple yes-no features such as open versus closed or rounded versus spread. Their proof-of-concept study used computer vision to track mouth area, height, and width in running speech from one speaker.

Even with smooth drift inside each vowel, the vowels fell into groups that a watcher could tell apart. Those groups did not always match the textbook feature charts, which are built from the speaker's side, in tongue and lip terms. The authors argue that measured lip geometry gives a truer picture of what a listener can actually use, and they test the vision tools themselves as part of the work.

Catalog record

Source: https://europepmc.org/article/MED/PMC13585966