A deep audio-visual speech enhancement network separates a speaker's voice using lip movements and voice representations, with self-enrollment to handle visual occlusions.
The network is trained on the MV-LRS [27], LRS2 [21], and LRS3 [22] datasets, and tested on LRS3
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2019 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
My lips are concealed: Audio-visual speech enhancement through obstructions
A deep audio-visual speech enhancement network separates a speaker's voice using lip movements and voice representations, with self-enrollment to handle visual occlusions.