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 receives three inputs: (i) the noisy audio to be en- hanced; (ii) the corresponding video frames; (iii) a reference audio containing speech from the target speaker
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.