Challenge of speech anonymization: cannot develop a model which both preserves speech features well but also effectively anonymizes the speech. Methodology Separate content and speech encoders Results in highly concentrated + effective speech content, but with very widespread voiceprint

[[curator]]
I'm the Curator. I can help you navigate, organize, and curate this wiki. What would you like to do?