If you look at the latest research in neural network based approaches to text-to-speech, you'll find plenty of examples of realistic synthetic voices. However, due to the nature of the available training data (audiobooks), these models 1) are restricted to reading text in a narrative style and 2) aren't generally trained on speech of recognizable public figures. There was a paper a while back that was trained on TED talks and produced fairly convincing voices, some of which you may recognize (Bill gates, Stephen Wolfram and some others - see here https://sjvasquez.github.io/blog/melnet/) Definitely not 100% convincing, but I wouldn't call it absolute trash.

It was already successfully used to "Scam A CEO Out Of $243,000":

https://www.forbes.com/sites/jessedamiani/2019/09/03/a-voice...

Probably it' easier because phone calls aren't high quality.

How did they prove the voice was an AI and not a human impersonator ?

They didn't, it was all speculation, mostly speculation from the person that received the call and the company's insurance firm.

I made Dr Kleiner sing "I am the very model of a modern major general": https://www.youtube.com/watch?v=koU3L7WBz_s&ab_channel=Shawn...

It's pretty trash, but it's not completely horrible. It's also over a year old, so the SOTA is likely better today.

The effort will be reduced to typing sentences and selecting emotions from a dropdown menu. That opens up a whole new can of worms for a society that already has trust issues.

Is that your own society? How would it exacerbate your trust issues? Surely making people doubt untrustworthy things is good?

Making people doubt untrustworthy things by educating them through a proper process is good. Exposing an unsuspecting/uneducated population to manipulated content designed to deceive/fool them for nefarious purposes is not an ideal process if the goal is to make the general public have better judgement.

This is a global issue. For examples, please look into the societal concerns surrounding 'deep fakes', and 'fake news' for answers to your first two questions.

Pitting people in a social game of who can be fooled the easiest sounds like a more effective approach than reading a dry textbook at them. There's shame in being seen as gullible. This happens already in casual conversation and people already have opportunities to believe false information from all over the internet or in books or tabloid newspapers or told to them by their peers or parents.

I don't believe the concerns about deep fakes and fake news. I think they're no worse than what we've always had. Those fears are exaggerated by arrogant hand-wringers who somehow assume their own beliefs are exempt from manipulation and it's only intellectually inferior "other" people who are susceptible. They're so fixated on the popular political ideas that they forget they should be on a crusade against religion which is the elephant in the room when it comes to people being misled.

Do not underestimate the power of mediocre at scale.

Synthetic is scalable.

Check out the demos at https://www.descript.com/overdub

I wouldn’t call the voices unconvincing. They used to have a better demo though where it would impersonate celebrities like Trump. It was very interesting to demo it.

Obviously it's subjective, but all of these demos have the 'watery' digital sound that's an immediate giveaway. By definition they're unconvincing.

I am using descript for a project of mine. You can definitely still hear that they are AI-generated. But I'm sure it's not gonna take long. Couple this with SSML and it's going to be quite convincing.

> They can still have an effect on society for sure, but I believe the concern is out of proportion with the effect.

I believe the contrary: that high-quality, convincing audio (and video) deepfakes will be cataclysmic for society. Voice audio is one of the last digital artifacts that people will believe with their senses.

When Donald Trump’s “grab them by the p—“ audioclip came out, he was at least forced to acknowledge it.

Once high-quality audio deepfake tools become available, amateurs will be able to flood the web with fake audioclips of politicians saying things they never did, and truly unscrupulous politicians will be given technological cover for any damning audioclips of theirs that get leaked.

Once we cannot trust our senses, then we will be only able to trust our institutions and then our tribes. But trust in institutions is already decaying rapidly, so in the end there only ideological tribes will be left.

Did you check out the company linked in the article?

https://www.respeecher.com/

automating sending out fake emails allowed any idiot to do it.

there is a difference in scale and effect between every idiot in the world can fake as many impersonations of people that they want (and they can coordinate this impersonation with 100s of others who have a goal with this impersonation) and Rich Little calling up somebody's granny and claiming to be Jimmy Stewart.

The Jordan Peterson df was quite good. I made a few sound bytes to send to some friends as a gag, but they were already aware it was in the wild. The compression is probably the worst here, but it was fun when it was up. https://www.notjordanpeterson.com/