AudioPrivacy: Parallel Audio Dataset for Speaker Profiling with Diverse Audio Types and Rich Attributes

Speech signals convey abundant speakerrelated metadata, yet current privacy research predominantly focuses on identitycentric voiceprint protection, leaving sensitive Speaker Attribute Privacy (SAP) largely underexplored. This paper introduces AudioPrivacy 1 , a large-scale Chinese dataset designed to systematically evaluate SAP leakage in realistic, everyday scenarios. Comprising 227.3 hours of audio from 1,000 speakers, it uniquely encompasses four parallel modalities: speech, singing, paralinguistic expressions, and nonvocal acoustic signals (e.g., footsteps). Annotated with 11 diverse attributes, including finegrained physiological traits often overlooked in traditional corpora, AudioPrivacy enables a granular analysis of acoustic privacy risks. Our evaluations reveal significant leakage across multiple attributes, even when inferred from non-vocal signals. Furthermore, we demonstrate that state-of-the-art Multimodal Large Language Models (MM LLMs) can precisely profile speakers and exacerbate these risks, underscores the urgent need to rethink privacypreserving mechanisms in the era of powerful audio foundation models.