Membership Inference Attacks Against Self-supervised Speech Models
This work addresses privacy concerns for users of widely deployed speech models, though it is incremental as it applies known attacks to a new domain.
The paper tackles the privacy risks of self-supervised speech models by applying membership inference attacks, showing these models are vulnerable with high AUC scores at utterance and speaker levels.
Recently, adapting the idea of self-supervised learning (SSL) on continuous speech has started gaining attention. SSL models pre-trained on a huge amount of unlabeled audio can generate general-purpose representations that benefit a wide variety of speech processing tasks. Despite their ubiquitous deployment, however, the potential privacy risks of these models have not been well investigated. In this paper, we present the first privacy analysis on several SSL speech models using Membership Inference Attacks (MIA) under black-box access. The experiment results show that these pre-trained models are vulnerable to MIA and prone to membership information leakage with high Area Under the Curve (AUC) in both utterance-level and speaker-level. Furthermore, we also conduct several ablation studies to understand the factors that contribute to the success of MIA.