Audio Deepfake Detector Flags Fake Parkinson Podcast: Ethics Row

Deep Fusion Films launched an AI podcast hosted by Michael Parkinson, the legendary British interviewer. Problem: Parkinson died in 2023. His cloned voice now interviews people he never met, with his family's permission but plenty of public unease.

How it worked

The producers licensed archive audio, trained a voice model on Parkinson's decades of broadcasts, and generated new interviews. The family signed off. The result sounds like Parkinson, asks questions like Parkinson, but Parkinson never said any of it.

Consent from the estate does not settle the ethics. Listeners hear a dead man's voice presenting new opinions. Guests answer questions from someone who is not there. Technically impressive and philosophically queasy. Call it a tribute if you like. It still puts words in a dead man's mouth.

Detection angle

An audio deepfake detector can distinguish the clone from archive recordings of the real Parkinson. Voice deepfake detection looks for the synthetic smoothness that even the best models leave behind. As AI resurrections multiply, deepfake voice detection becomes the only way to know whether you are hearing history or software.

Here is the blunt version: the dead cannot consent to new opinions. Estates can sign papers, but the person is gone. Label AI voices clearly, always. If a podcast does not tell you the host is synthetic, that is deception, permission slip or not. The family consented. The audience was not told clearly enough. Those are different things.

Dead voices should stay dead, or be labelled. Try the AI Detector on suspicious audio.