A Comparative Study of Speech Anonymization Metrics

Mohamed Maouche, Brij Mohan Lal Srivastava, Nathalie Vauquier, Aurélien Bellet, Marc Tommasi, Emmanuel Vincent

Speech anonymization techniques have recently been proposed for preserving speakers’ privacy. They aim at concealing speakers’ identities while preserving the spoken content. In this study, we compare three metrics proposed in the literature to assess the level of privacy achieved. We exhibit through simulation the differences and blindspots of some metrics. In addition, we conduct experiments on real data and state-of-the-art anonymization techniques to study how they behave in a practical scenario. We show that the application-independent log-likelihood-ratio cost function Cminllr provides a more robust evaluation of privacy than the equal error rate (EER), and that detection-based metrics provide different information from linkability metrics. Interestingly, the results on real data indicate that current anonymization design choices do not induce a regime where the differences between those metrics become apparent.

 DOI: 10.21437/Interspeech.2020-2248

Cite as: Maouche, M., Srivastava, B.M.L., Vauquier, N., Bellet, A., Tommasi, M., Vincent, E. (2020) A Comparative Study of Speech Anonymization Metrics. Proc. Interspeech 2020, 1708-1712, DOI: 10.21437/Interspeech.2020-2248.

  author={Mohamed Maouche and Brij Mohan Lal Srivastava and Nathalie Vauquier and Aurélien Bellet and Marc Tommasi and Emmanuel Vincent},
  title={{A Comparative Study of Speech Anonymization Metrics}},
  booktitle={Proc. Interspeech 2020},