Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Voice Privacy from an Attribute-based Perspective

Mehtab Ur Rahman, Martha Larson, Cristian Tejedor-Garcia

Abstract

Voice privacy approaches that preserve the anonymity of speakers modify speech in an attempt to break the link with the true identity of the speaker. Current benchmarks measure speaker protection based on signal-to-signal comparisons. In this paper, we introduce an attribute-based perspective, where we measure privacy protection in terms of comparisons between sets of speaker attributes. First, we analyze privacy impact by calculating speaker uniqueness for ground truth attributes, attributes inferred on the original speech, and attributes inferred on speech protected with standard anonymization. Next, we examine a threat scenario involving only a single utterance per speaker and calculate attack error rates. Overall, we observe that inferred attributes still present a risk despite attribute inference errors. Our research points to the importance of considering both attribute-related threats and protection mechanisms in future voice privacy research.

Voice Privacy from an Attribute-based Perspective

Abstract

Voice privacy approaches that preserve the anonymity of speakers modify speech in an attempt to break the link with the true identity of the speaker. Current benchmarks measure speaker protection based on signal-to-signal comparisons. In this paper, we introduce an attribute-based perspective, where we measure privacy protection in terms of comparisons between sets of speaker attributes. First, we analyze privacy impact by calculating speaker uniqueness for ground truth attributes, attributes inferred on the original speech, and attributes inferred on speech protected with standard anonymization. Next, we examine a threat scenario involving only a single utterance per speaker and calculate attack error rates. Overall, we observe that inferred attributes still present a risk despite attribute inference errors. Our research points to the importance of considering both attribute-related threats and protection mechanisms in future voice privacy research.
Paper Structure (17 sections, 2 figures, 6 tables)

This paper contains 17 sections, 2 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work and Background
  3. Uniqueness Analysis for Attribute-based Profiles
  4. Attacker Specification for the Re-identification Attack
  5. Experimental Setup
  6. Dataset
  7. Method for Analyzing Speaker Uniqueness
  8. Method for Attribute-based Re-identification Attack
  9. Setup for Attribute Inference
  10. Results of Analysis and Attack
  11. Analysis of Speaker-Level Attribute Inference
  12. Analysis of Utterance-Level Attribute Inference
  13. Analysis of Inference on Anonymized Utterances
  14. Re-Identification Attack
  15. Conclusion and Outlook
  16. ...and 2 more sections

Figures (2)

  • Figure 1: Speaker singling out under original and anonymized single utterance per speaker setting. Each dot represents the percentage of speakers whose anonymity set size satisfies the given threshold in one independent run.
  • Figure 2: Impact of attribute inference on the uniqueness across speakers. For different anonymity thresholds $k$, the figure shows the proportion of speakers whose anonymity set increased in size (reduced risk), decreased in size (higher risk), or remained unchanged relative to ground truth labels.