DDPO-VC: Speaker De-Identification via Diffusion Denoising Policy Optimization
For researchers in privacy-preserving speech processing, this work provides a method that maintains utility (e.g., cognitive health assessment) while de-identifying speakers, overcoming limitations of disentanglement approaches.
DDPO-VC addresses the challenge of balancing privacy and utility in speaker de-identification, where disentanglement methods fail due to correlations between privacy and utility variables. Using reinforcement learning post-training with diffusion models, it outperforms strong baselines in both privacy preservation and cognitive utility on dementia speech benchmarks.
A key challenge of speaker de-identification is the balance between privacy and utility. Many utility variables, such as the cognitive health status of the speaker, are correlated with the privacy variable, such as the speaker identity, violating the independence assumption held by the disentanglement-based approaches, causing leakage of private information and the loss of useful information for downstream tasks. To tackle this challenge, we propose a general framework, DDPO-VC, for speaker de-identification through reinforcement learning-based post-training with diffusion models. Learning from reward signals combining knowledge from privacy-focused and utility-focused teachers, our method outperforms various strong \deid/ methods in both privacy preservation and cognitive utility on two commonly used dementia speech benchmarks. Please check out our code\footnote{\href{https://github.com/cactuswiththoughts/DDPO-VC}{https://github.com/cactuswiththoughts/DDPO-VC}} and demo\footnote{\href{https://cactuswiththoughts.github.io/SpeakerDeID-Demo/}{https://cactuswiththoughts.github.io/SpeakerDeID-Demo/}}.