VoxPrivacy Benchmark for Speech Language Model Interactional Privacy
September 4, 2026
VoxPrivacy is a new benchmark designed to evaluate interactional privacy in speech language models (SLMs). It specifically measures whether a model can distinguish between users to prevent the leakage of one user's contextual sensitive information to another in multi-user environments.
HOW THIS AFFECTS YOU
●
researcherThis introduces a necessary metric for speaker-aware response adaptation in SLMs.
●
policyThis defines a critical safety requirement for deploying voice assistants in shared household settings.