The paper introduces Varys, a modular and extensible testbed for automating large-scale interactions with voice interfaces, enabling reproducible collection and analysis of network traffic, audio, transcripts, and metadata. Using the dataset collected with Varys, we also reproduce and evaluate two existing voice-fingerprinting approaches.
 

Citation: Shiva Parsarad, Anastasiya Merkushova, Milan Van Zanten, and Isabel Wagner. 2026. Varys: A Testbed to Automate Interactions with Voice Interfaces. In Proceedings of the 21st International Conference on Availability, Reliability and Security (ARES), Linköping, Sweden. [PDF available]

Repository for the Varys tool: https://gitlab.com/dmi-pet-public/vanzanten2026varys

Repository for voice fingerprinting based on data collected with Varys: https://gitlab.com/dmi-pet-public/parsarad2026voicefingerprinting

Dataset of interactions with voice assistants, collected with Varys: https://zenodo.org/records/17878434