VeriSpoof: Unified Audio LLM for Deepfake Forensics
November 2, 2026
VeriSpoof is an audio language model that unifies five forensic tasks, including deepfake detection and generator attribution. It provides grounded, verifiable rationales for its decisions, allowing analysts to verify claims against waveform measurements or metadata.
HOW THIS AFFECTS YOU
●
builderYou can integrate multi-task audio forensics into a single LLM-based pipeline.
●
policyThis provides a more transparent and verifiable method for detecting synthetic speech.