Jean-Philippe Thiran, Mihai Gurban
We present a method for dynamically integrating audio-visual information for speech recognition, based on the estimated reliability of the audio and visual streams. Our method uses an information theoretic measure, the entropy derived from the state probab ...
2008