Yikes: Now deepfakes are getting the voices right too

Yikes: Now deepfakes are getting the voices right too

A late-night palate cleanser that operates more as a proof of concept than as a polished final product. Nothing in this clip from deepfake maestro Ctrl-Shift-Face works perfectly, as you’ll see below. Trump’s visage looks too young and glowy layered on top of Bob Odenkirk’s, and there’s no way for a human alive in 2019 to believe in the authenticity of that face minus the president’s famous coif above it. The voice trick has bugs in it too. The cadence is obviously unnatural at points. The audio doesn’t always sync to the character’s lips.


That is, of course, unmistakably Trump’s face. And it also is, unmistakably, Trump’s voice. This is the first time I’ve heard an audio deepfake and I’m surprised at how good it is. I shouldn’t be — it must be easier to fake sound than to fake visuals. Getting the inflections right in a particular stretch of dialogue to mimic natural speech will be a challenge, but the most basic technological task, breaking down the unique sound of a particular voice and using the component parts to build new audio, has already been completed.

We really are right on the doorstep of substituting one person in all their particulars for another in A/V media and having it seem perfectly real.

I assume the next step, then, is replacing an entire head. As strange as it is to see Trump’s face in a “Better Call Saul” monologue here, you’re never unaware that that’s still *mostly* Bob Odenkirk. Imagine how it’ll feel when they can swap in the entire look — Trump’s head, his build, the trademark blue suit with red power tie. Witchcraft.

Trending on HotAir Video