ReVISE: Self-Supervised Speech Resynthesis with Visual Input for Universal and Generalized Speech Regeneration

Keyphrases

• Regeneration 100%
• Visual Input 100%
• Visual Speech Recognition 100%
• Speech Resynthesis 100%
• Self-supervised Speech 100%
• Audio 66%
• Audiovisual Speech 66%
• Text-to-speech Synthesis 66%
• Superior Performance 33%
• Discrete Units 33%
• Training Data 33%
• Speech Quality 33%
• Initialize 33%
• Quality Model 33%
• Single Model 33%
• Inpainting 33%
• Speech Synthesis 33%
• Suppress Noise 33%
• In-the-wild Videos 33%
• Audiovisual Synchrony 33%
• Acoustic Conditions 33%
• Auditory Distortion 33%
• Speech Models 33%
• Image-to-speech 33%
• Self-supervised Speech Models 33%

Computer Science

• Speech Recognition 100%
• Speech Synthesis 100%
• Text To Speech 66%
• Superior Performance 33%
• quality model 33%
• Training Data 33%
• Video Inpainting 33%
• Synchronism 33%