Live data from Hacker News

They stole my voice with AI

jeffgeerling.com

431–440 of 461 posts

Re: They stole my voice with AI

#431

Earlier quoted context omitted.

That's convenient, right? AI will be more than perfect and also be able to be imperfect if you want it, it'll just do anything you want. Here's a better prediction: Almost no one wants to listen to whatever low effort trash gets pumped into an advanced TTS.

You wouldn't even be able to tell at the end of the day, that's the point they're making.

Bold claim. To be frank, I'll believe when I see it (or hear it).

My prediction is that to get to that point you need to hire so many editors that you may as well have just hired the VA outright. We'll get more fronts in the meantime not too dissimilar from the Amazon Checkout debacle.

Re: They stole my voice with AI

#432

Ex-Google CEO says successful AI startups can steal IP and hire lawyers to ‘clean up the mess’ / “But if nobody uses your product, it doesn’t matter that you stole all the content,” Eric Schmidt said during a recent talk at Stanford that has been taken offline. Since that guy was CEO of Google it’s all good right??? https://www.theverge.com/2024/8/14/24220658/google-eric-schm...

well that does nail down one huge societal issue that multiple domains need to address: fine and punishments are so low that they are simply an expense instead of a deathblow.

We definitely need to overhaul a lot of these white collar fines. I just watched a video today and learned the federal maximum fine for being caught using child labor is capped at $15k per worker. No wonder child labor has skyrocketed over the decade.

Re: They stole my voice with AI

#433
post #419

how do people know if its a very similar sounding voice, or identical?

Not to be crude, but they have ears. and you'll hear uncanny likeness of a voice you follow everyday when it comes up. It's kind of like saying "how do we know that face is similar/identical"? Humans are surprisingly good at knowing when something feels off in other humans. Even if we lack the vocabulary to fully explain the difference.

The problem is that there are a lot of humans, and a lot of humans with online presence. I wouldn’t really be confident in identifying a „stolen face“ online because you’re bound to look similar to someone when creating a face. Does that mean that the look was stolen or is it just coincidence? What is the likelihood of a random AI voice sounding similar to some real online creator?

Re: They stole my voice with AI

#434
post #266

[flagged]

I'm in the camp that thinks celebrities take way too much mental space of the regular person on the internet. It distorts ones ability to differentiate as there are these huge gravity centers that everybody makes comparisons to, even subconsciously. If Scarlett Johansson wasn't famous and Sam Altman didn't offer her the job, everybody would have no trouble in recognizing the differences in the voice. Priming is an ex…

you're being a bit misleading here by implying that humans have perfect voice recognition to begin with.

Try to go to a foreign language and figure out the difference between 10 famous voices. Odds are at least half of them will blend to you. Becuase you do not spend everyday listening to that language.

Now try and find a person who has a similar voice to your parent, then ask 10 random people to see if they can tell the difference. I'm sure 8/10 wouldn't be able to. I'm sure you can, though.

Voice recognition is a mix of factors, which includes familiarity and language processing. If Scarlet wasn't famous, it wouldn't have been as huge an issue because not enough fans would be familiar enough. Friends would still complain though.

>and Sam Altman didn't offer her the job

Handwaving this doesn't help. That's the entire basis of Ford v. Milton that made Ford lose the case. You can't just go behind a VA's back once they decline and use their likeness anyway. Even outside of law, that's just scummy.

Re: They stole my voice with AI

#435

Earlier quoted context omitted.

They reached out to Scarlett Johanssen. She said no. So they found a soundalike and Tweeted out references to the movie Her (starring Scarlett Johanssen as an AI chat bot) in the days leading up to launch. Scummy as fuck from OpenAI regardless of the technical legal rights and issues involved.

What's scummy about it? If you want a specific kind of voice and approach one person about it but she doesn't agree what's wrong with asking another one? If there were just two blonds in the world, one famous and the other not and you wanted a blonde actress for the role and she said no is it scummy to hire the other one? Is it scummy to hire "discount" Matt Damon instead of Matt Damon?

>Is it scummy to hire "discount" Matt Damon instead of Matt Damon?

If Matt Damon said no (e.g. leaving a paper trail) and you want people to pretend it's Matt Damon anyway, yes. Look up Ford v. Milton. You can't just go behind an artist's back like that.

If you want to argue about soundalikes in spinoffs, those are discussed in contracts (and those kinds of figures are making 7+ figures as is, so it's fair on them to choose to let them do that).

Re: They stole my voice with AI

#436
post #83

Earlier quoted context omitted.

I strongly disagree. I don't know the rights around one's own voice, but the idea that you suddenly lose ownership of something because you shared it online is the exact thing that many people take issue with when it is written in the terms of service for social networks, creator tools (adobe), etc.

I didn't mention ownership and I don't think you should lose it (nor does one lose it really even in this case, legally). But I do think that in cases like these, where there's money involved and YouTube, that they should have the means to prevent it.

Issue is that Youtube profits off it while signing away liability. There's no incentive to prevent it.

Re: They stole my voice with AI

#437

Earlier quoted context omitted.

Ownership as a whole is a social/legal construct, no? [1] You only "own" something insofar as you control it. Societies build frameworks that aid people in controlling things based on who they think deserves control of those things. This clearly can vary widely between societies; the USA used to think owning people was perfectly OK and some places still do. Nobody has absolute control over (almost?) anything, and I t…

Many social/legal constructs are for lack of a better term, idiotic. Step back for a second and think about what ownership actually implies? Exclusive control over a scarce resource. If a resource isn't scarce, where your control over it doesn't impede or restrict other peoples control over it, there is no feasible model that explains why one person has control or ownership of it over another. Social/legal constructs…

>Step back for a second and think about what ownership actually implies?

It implies identity, and control. The former being most relevant here. And yes, if you want to use someone's identity you need to at least get consent.

>Exclusive control over a scarce resource.

"Scarce" is doing a lot of heavy lifting. And in the grand scheme, everything can be "scarce". The internet isn't infinite either, only in a virtual sense (no pun intended).

Meanwhile, thoughts and data are resources, and you can own the idea behind it. Hence, "intellectual property".

> there is no feasible model that explains why one person has control or ownership of it over another.

That's where identity came in. If an inventor got run over tomorrow, there may still be iterations on the invention, but it would completely alter the course of iteration. That's ownership in a sense. Ownership that your ideas can't just be understood or replicated unless you transfer your thoughts to another medium. And like any conversion, data is lost in the exchange.

>they exist to give privilege and power to certain groups at the expense of others.

Yeah, the less off inventors. Do you know why IP is automatically granted? Because someone unaware of the legal channels would just have their own IP signed off once anyone in business gets a whiff of it.

of course, the priveledge still abuse it. But when everything becomes a free for all, the rich get richer.

Re: They stole my voice with AI

#438

Earlier quoted context omitted.

Not to be crude, but they have ears. and you'll hear uncanny likeness of a voice you follow everyday when it comes up. It's kind of like saying "how do we know that face is similar/identical"? Humans are surprisingly good at knowing when something feels off in other humans. Even if we lack the vocabulary to fully explain the difference.

The problem is that there are a lot of humans, and a lot of humans with online presence. I wouldn’t really be confident in identifying a „stolen face“ online because you’re bound to look similar to someone when creating a face. Does that mean that the look was stolen or is it just coincidence? What is the likelihood of a random AI voice sounding similar to some real online creator?

>I wouldn’t really be confident in identifying a „stolen face“ online because you’re bound to look similar to someone when creating a face. Does that mean that the look was stolen or is it just coincidence?

Luckily, national ID can easily solve that problem offline. It'd be a more interesting problem if some company searched far and wide to identify someone that has a similar face to [famous person]. And that's part of the issue in that there's no way to verify "is my voiced sampled here?". Opening the black box can help with that, but that's something companies are fighting to the death to not allow.

> What is the likelihood of a random AI voice sounding similar to some real online creator?

Given how much data was scraped, less likely than the fact that their data is in fact in the training set.

Re: They stole my voice with AI

#439
post #345

Earlier quoted context omitted.

Changing an image timestamp -- as in, exif metadata -- is trivially easy for anyone with a computer or phone, a quick search will tell you how to do it on any device with no skill involved. There's a difference between "easy for a small percentage of people" (modify the content of a photo in an undetectable way, find a software vulnerability, etc.) and "only a small percentage know how to do it" (modify exif values,…

The problem that to make a right query, one has to know at least part of the answer. To search for exif modification tool, it's essential to understand: - what is image file, and how information stored in it. - the concept of `metadata`, and exif in particular. - the concept that any file can be modified in place, or when you make a copy. - know what where's ecosystem of open/free/custom tools that can do an unusual…

This.

I'm currently the originator and lead on a project that will have huge impact on our bottom line. The team is mostly junior people, so when I presented the plan the response I got wasn't "I didn't think we could do that", it was "I didn't even know that that is a thing".

There used to be a joke about Rumsfeld's "known unknowns" vs. "unknown unknowns", but it's so true. And to the vast majority of people in the world the question isn't "how" to change a timestamp, it's understanding that a timestamp can even be changed.

Re: They stole my voice with AI

#440

[flagged]

> Scarlett's baseless accusation https://www.theguardian.com/film/article/2024/jul/18/scarlet... > The actor made headlines earlier this year when she issued a public statement saying OpenAI’s chief executive, Sam Altman, had contacted her in September 2023 to provide the voice for Sky, in an ode to her role in the Spike Jonze film Her, but she had declined for “personal reasons”. In May, as the company released Sky,…

This is not evidence, and it is still baseless. They literally hired a completely different voice actress. Nothing else needs to be said. What the fsck does it matter if it was "similar to" Scarlett Johannsen, even if it was (which it wasn't)? Should an actor be able to sue another actor who looks similar because of the work they got?
Post reply on HN