VeriSpoof-Base broadens speech deepfake forensics

VeriSpoof-Base moves speech anti-spoofing beyond a brittle real-or-fake score toward a more useful forensic workflow.

• –One model handles detection, generator identification, source verification, localization, and channel-condition analysis.
• –Its explanation-first design makes predictions more auditable than opaque binary classifiers.
• –The reported 4.43% pooled EER and #4 ranking among 28 systems are strong, though they remain benchmark results rather than courtroom-grade evidence.
• –Local inference lowers privacy and deployment barriers, but the CC BY-NC 4.0 license limits commercial use.
• –Developers should validate performance on their own languages, codecs, attacks, and real-world recordings before production use.