Welcome to Lip-reading demo!
This web-app ilustrates possibility of using pre-trained Deep Neural Networks on webpages,
using only client browser.
It will capture your pronounced word using webcamera,
and will try to predict spoken word using only image sequence.
Project is in testing phase.
Setup
- Please use Google Chrome
- Enable Camera in top-left corner
- Ensure you have enough light, camera is directly facing you,
and you have at least 25 FPS (green FPS indicator)
- Otherwise, prediction will not work
How to use it
- Wait for CLM Tracker to find your face and match it
- When ready, press SPACEBAR, you'll see red border in video window,
this means program will record next 1 second of frames as soon as you open mouth
- You'll see predicted distribution of possible words on the right
Sources