Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Comparing human and neural network lip readers.

B P Yuhas1, M H Goldstein

  • 1Johns Hopkins University, Baltimore, Maryland 21218.

The Journal of the Acoustical Society of America
|July 11, 1991
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

The Influence of Claude Bernard on Physiological Concepts.

Annals of medical history·2021
Same author

Bilateral subperiosteal orbital hemorrhages after a compressive thoracoabdominal injury.

The Journal of trauma·2001
Same author

Fungal ring infiltrates in disposable contact lens wearers.

The CLAO journal : official publication of the Contact Lens Association of Ophthalmologists, Inc·2001
Same author

Fungal keratitis in a daily disposable soft contact lens wearer.

The CLAO journal : official publication of the Contact Lens Association of Ophthalmologists, Inc·2001
Same author

Facial nerve injury following superficial temporal artery biopsy.

Dermatologic surgery : official publication for American Society for Dermatologic Surgery [et al.]·2001
Same author

Diagnosis of hemophilia made after intraoperative bleeding during attempted penetrating keratoplasty in an elderly patient.

The CLAO journal : official publication of the Contact Lens Association of Ophthalmologists, Inc·2001
Same journal

Interaction of near-wall bubble arrays with acoustic waves induced by an oscillating rigid wall.

The Journal of the Acoustical Society of America·2026
Same journal

Ultra-broadband underwater acoustic projector based on transverse resonance orthogonal beam (TROB) mode and acoustic matching layer technique.

The Journal of the Acoustical Society of America·2026
Same journal

Fine-scale quantitative analysis of bowhead whale (Balaena mysticetus) song shows varying stability of song types.

The Journal of the Acoustical Society of America·2026
Same journal

High-resolution depth estimation for multiple wideband sources in deep sea via sparse Bayesian learninga).

The Journal of the Acoustical Society of America·2026
Same journal

Depression markers in speech: An approach based on tract variables dynamics.

The Journal of the Acoustical Society of America·2026
Same journal

The oyster toadfish (Opsanus tau) alters active and diurnal calling amid vessel noise in New York City.

The Journal of the Acoustical Society of America·2026
See all related articles

Neural networks and humans perform equally well in identifying vowels from lip images. This study compares their accuracy in visual vowel recognition, finding similar correct identification rates for both.

Area of Science:

  • Computer Vision
  • Human Perception
  • Speech Recognition

Background:

  • Visual speech recognition, or lip-reading, is a complex task.
  • Neural networks are increasingly used for pattern recognition tasks, including image analysis.
  • Comparing artificial intelligence performance to human capabilities provides valuable insights.

Purpose of the Study:

  • To experimentally compare the performance of neural networks and human subjects in identifying vowels from static images of a speaker's mouth.
  • To quantify the accuracy of visual vowel identification for both systems.

Main Methods:

  • Static images of a speaker's mouth were presented.
  • Both neural networks and human subjects were tasked with identifying nine distinct vowels.
  • The rate of correct identifications was recorded and analyzed for each vowel and overall.

Related Experiment Videos

Main Results:

  • Neural networks and human subjects demonstrated comparable overall rates of correct vowel identification.
  • Performance was similar for both groups on a vowel-by-vowel basis.
  • The probability of correct identification and 95% confidence limits were calculated for each vowel.

Conclusions:

  • Neural networks achieve human-level performance in visual vowel identification from static lip images.
  • The findings suggest that AI can effectively replicate human capabilities in this specific aspect of speech perception.
  • Further research can explore dynamic mouth movements and more complex speech elements.