Reducing audio membership inference attack accuracy to chance: 4 defenses

TL;DR

This paper explores privacy vulnerabilities in speaker identification models, demonstrating high attack accuracy and proposing defenses like prediction obfuscation and adversarial training to reduce it to chance levels.

Contribution

It generalizes membership inference attacks to audio speaker identification and evaluates defenses to mitigate privacy risks.

Findings

Attack precision up to 85.9% on LibriSpeech

Defense methods reduce attack accuracy to chance

Effective defenses include prediction obfuscation and adversarial training

Abstract

It is critical to understand the privacy and robustness vulnerabilities of machine learning models, as their implementation expands in scope. In membership inference attacks, adversaries can determine whether a particular set of data was used in training, putting the privacy of the data at risk. Existing work has mostly focused on image related tasks; we generalize this type of attack to speaker identification on audio samples. We demonstrate attack precision of 85.9\% and recall of 90.8\% for LibriSpeech, and 78.3\% precision and 90.7\% recall for VOiCES (Voices Obscured in Complex Environmental Settings). We find that implementing defenses such as prediction obfuscation, defensive distillation or adversarial training, can reduce attack accuracy to chance.