Publications
Analysis of embedding-based emotional preservation metrics for voice conversion models
By Théo Nguyen, Tom Bäckström, Rainer Martin
Published in: Odyssey 2026 — 04/2026
Emotional content is often degraded in anonymization models. To quantify the loss of emotional information, benchmarks rely nowadays on hard label matching, which is highly constraining. We propose ways of getting rid of these limitations.
Evaluating voice anonymisation using similarity rank disclosure
By Shilpa Chandra, Matteo Pettenò, Nicholas Evans, Michele Panariello, Massimiliano Todisco, Tom Bäckström, Dorothea Kolossa, Rainer Martin, Themos Stafylakis, Nicolas Gengembre
Published in: Odyssey 2026 — 30/04/2026
This work uses similarity rank disclosure (SRD) to evaluate voice anonymisation, providing an information-theoretic assessment of privacy.
Privacy in Speech Technology
By TOM BÄCKSTRÖM, Senior Member IEEE
Published in: Proceedings of the IEEE
This article presents a tutorial overview of privacy in speech technology that covers a wide range of threats, methodologies, and algorithms.