TL;DR of the paper, generated automatically.

An open audio model that spots fake speech and explains its call.

• Detection: 2.14% EER pooled over 14 test sets, third of 28 systems.
• Training: updates 29.7M of 2.85B parameters, fits one 24 GB GPU.
• Releases: both models, inference code, a demo, and VeriSpoof-Bench.