Study reveals what people see when they read lips
3 days ago
- #visemes
- #speech-perception
- #lip-reading
- New research from the University of Kansas uses network science to study lip-reading errors, focusing on visual characteristics (visemes) rather than sounds.
- A visual map of about 20,000 English words revealed that people often confuse words with common look-alikes; about one-third of words resemble at least one other.
- Errors are non-random and more likely when visually similar words are close in the visual network, making some words harder to lip-read.
- Findings suggest lip-reading accuracy is lower than assumed, with mistakes typically being only one or two visemes off.
- Applications include improving lip-reading training by tracking error reduction and enhancing AI transcription systems (e.g., Zoom) by integrating visual data.
- Future research aims to explore machine-learning applications and assistive technologies for speech understanding.