If you hear podcasting star Joe Rogan say something dumb, it may not be his fault – an AI has cloned his voice

Here’s another one of your regular reminders that AI software can be creepy. Engineers at Dessa, an AI startup focused on helping enterprises use machine learning, have managed to clone the voice of Joe Rogan, the host of the popular podcast show The Joe Rogan Experience. Dessa called it “the most realistic AI simulation of a …

[JetSetJim]

It's very good, for sure, but I think it's still in the uncanny valley. Doesn't quite work for me, but I'm vaguely familiar with his normal cadence and obviously am biased as I knew it was ai when listening to it

Saying that, they'll probably now reveal it was an actual recording of Joe as a prank/commentary on modern AI research

[DavCrav]

Sounded like a normal person to me, but then I've never heard of whoever this famous person is. And I'm not American, so I'm less attuned to their idiosyncrasies.

[Anonymous Coward]

Wouldn't be the first time!

It's a good few years ago now, but El Reg once reported on the 'Mosquito' teenage repellant almost-ultrasonic device, the idea behind it being that only the young could hear the very top end of the audio range and would be sufficiently irritated to move along

The article had a link to an mp3 file, allegedly of the sound of the thing, for those with sufficiently good hearing, However, looking at the file in both Wavelab and Audacity showed it to consist only of silence.

FWIW, to my ears this pseudo Joe Rogan is very close. Knowing in advance that it was 'fake' I could spot that the forceful projection of personality of the original seemed missing, but if I hadn't known in advance, I'd probably have taken it as him just being a bit distracted.

Rather worrying, though hardly unexpected.

[Graham Dawson]

Re: Wouldn't be the first time!

I still hear those mosquito things. Guess I'm one of the yoof, despite my wife's claims to the contrary.

[regregular]

It is slightly uncanny, but I believe it is a kind of "you knew before, so you were concentrating on the bits that sound off".

If this had been clipped into his regular podcast right after the commercials and before the guest intro I would not have batted an eye This tech is getting scary good.

Also, chimps vs humans in hockey sounds like an incredible idea and is no less stupid that WWE or cage fights. I'd watch the shit out of that.

[JeevesMkII]

It had the feel of a human reading an unrehearsed speech from notes in front of them, but if you told me that it was a human doing the reading I'd probably have believed it.

We're a few years away, but voice actors are on serious notice that their jobs are going away real soon now and movie/television actors can't be far behind when photorealistic animation becomes cheaper to produce.

[Rich 11]

If you hear podcasting star Joe Rogan say something dumb

...you may not be the least bit surprised.

[Winkypop]

Re: If you hear podcasting star Joe Rogan say something dumb

^^^ This

[DavCrav]

The perfect time for HMRC and the banks to start rolling out 'my voice is my password' identification.

[cornetman]

I've watched a few Joe Rogan shows so I'm fairly familiar with his voice and I have to say I was very impressed.

If you listen very carefully, there are some artifacts in the voice that make it sounds a little artificial.

However, consider that most people might put that down to compression artifacts in the sound track, I think you could probably get away with it.

I was pretty impressed with the inflection as well. Seemed appropriate to the position of the sentence and consistent with the flow.

Scary and it can only get better. :(

[Graham Dawson]

It'd be nice to know how much manual tweaking went into it after it was generated, if any.

[EveryTime]

I'll also toss a vote into the 'there are artifacts, but they could be from compression' category.

It sounds like silence compression is changing the word spacing slightly, along with a compression re-start on each word.

On a related topic, I've been getting a lot of "handwritten" advertising mail lately. It's quite convincing at first glance, with variations in letter shapes, spacing and line slope. Those were all features that made it trivial to identify earlier machine "handwriting". I expect that it's only a little more work to further process the voice output to get rid of the more obvious "tells".

[TopCat62]

And not just audio...

It's not just audio... it's going to be video as well.

https://www.ted.com/talks/doug_roble_digital_humans_that_look_just_like_us