DP-VOXLET: Provable Speaker Anonymization for Disentangled Speech Representations
2026-08-31 • Cryptography and Security
Cryptography and Security
AI summaryⓘ
The authors focus on systems that hide who is speaking in an audio clip while keeping the message and tone the same. They introduce a new way to protect speaker identity using a strict privacy method called differential privacy, which guarantees limits on how well someone can guess who the speaker is. Their method works with existing speech models and improves the balance between keeping the speaker anonymous and preserving audio quality. This formal approach goes beyond previous methods that were more guesswork-based.
speaker anonymizationsemantic contentprosodydifferential privacydisentangled representationsprivacy guaranteeequal error rateadversaryre-identification
Authors
Ivoline Ngong, Jack D'Iorio, Hailey Schoppe, Christopher Liberatore, Nichole Schimanski, Taisa Kushner, Joseph P. Near
Abstract
Systems for speaker anonymization obfuscate the speaker of an utterance, while maintaining its original semantic contents and prosody. Recent solutions for speaker anonymization rely on learned representations that disentangle an utterance into semantic contents and speaker properties. To anonymize an utterance, these systems replace the speaker properties while leaving the semantic contents unchanged---an approach that can produce strong results on empirical measures of privacy. In this work, we introduce speaker differential privacy, a formal definition of speaker anonymization based on the framework of differential privacy, and a mechanism for speaker anonymization that provably satisfies the definition. In contrast to prior heuristic-based anonymization systems, our approach enables a provable lower bound on re-identification success rate (e.g. equal error rate) for any possible adversary. We implement our approach in a framework that is compatible with existing disentangled representations. Compared to the prior work on differential privacy for speaker anonymization, our approach achieves significantly higher utility.