Uh?? How can any voice sound /too much/ like Scarlett Johansson?
:)
If they didn't and just cloned her voice, it's more disregard for creators and artists than I would have thought possible. What were they thinking?
Edit after reading the official story... not sure I believe it, seems disingenuous, at best they chose someone because they really really sounded like Scarlett Johansson, and no one said, it might be a problem.
https://openai.com/index/how-the-voices-for-chatgpt-were-cho...
Edit: I know it's not really Scarlett Johansson, but it does sound very similar, and they can do better to make it right, in terms of sound and legality!
In the case of computer-generated voices, there are qualities that are desirable that also happen to be attributes of a real person's voice. How many of these desirable attributes can a computer-generated voice have before it's considered too close to the set of attributes a particular person's voice has?
Her - https://www.youtube.com/watch?v=Ij0ZmgG6wCA
If it was a soundalike then it would be a legal issue obviously, nothing to discuss.
Looking at the HN comments they are told it sounds like Her so that's what they believe. So you can't trust the NPC's to decide, they just regurgitate media headlines.
How do you quantify it?
This is the case for all mobile and web apps and has been the norm for over a decade now. If you want control over the UI or functionality, use local software that doesn't check some server for feature flags.
Even if Scarlett Johansson would agree to license her voice, which is a big if, it's not her voice that's being removed. It's a different actress that some users thinks sounds like Scarlett.
https://openai.com/index/how-the-voices-for-chatgpt-were-cho...
Next we are getting Ex Machina?
if your product relied on temu scarjo, you never had a product.
“Florida person finds loophole in system to marry ChatGPT voice assistant”
“ClosedAI back peddles controversial decision to ‘lobotomize’ voice assistance after widespread dislike”
“It is now legal to marry 3D printed humans with AGI in these 25 states”
“Florida person initiates divorce with ChatGPT, claims 50% stake of OpenAI. Florida state courts open to it. OpenAI tumbles at market open by 11%”
I don’t fully recall the ending but doesn’t the AI grow past the guy and “break up” with him, leaving him devastated at the end?
A bit sounds like the Replika AI drama from the last year. </SPOILER>
Voices such as Shatner or Walken are at least as much about speech patterns than the voice, giving you another axis to compare voices against, so I can somewhat see those as being trademarkable. But when I hear the ChatGPT voice, it just sounds like "slightly-flirty generic female voice 03" to me.
It was left a little in doubt whether the AI really did reach 'enlightenment' and beam itself to the stars, or the company/government shut it down because society was collapsing.
Real question.
Is there some waveform comparison that a court would accept?
https://x.com/sama/status/1790075827666796666
> her
And this one:
https://x.com/prafdhar/status/1790789900650037441
> @alex_conneau: came up with the vision of HER before anyone at OpenAI had, and executed relentlessly!
I'm not saying it's her voice. I'm not disputing that they found someone with a similar sounding voice. But OpenAI should still aim to license from Scarlett Johansson and fine-tune on real licensed audio from her, the real deal, and let users switch to it if they want to.
So for me, this voice had no impact — sure I noticed it seemed a bit "flirty", but that's not a thing that engages me in any way as it feels equally fake when a human does it, and if anything I pattern-matched to the Pierson's Puppeteers in Ringworld; the original Alexa advert was moderately creepy, but I could see they were trying to mimic the computer in Star Trek; but one example I do have of being disturbed by a product advert was the use of a cheerful up-beat soundtrack for "The Robot Dog With A Flamethrower | Thermonator": https://www.youtube.com/watch?v=rj9JSkSpRlM
However, I did prefer Sky above the other voices. It didn't come across as seductive or flirty to me, just something that sounded close to the Google Assistant or Alexa. "Neutrally approachable" is how I'd put it, like a good receptionist, a mix of casual warmth and surface level concern. Shrug. I hope they bring her back soon.
I do not see how you could interpret the ending of Her this way.
More frightening is Ex Machina, which shows what happens when such an AI isn't regarded as a person by its creator, and sees fit to take personhood for itself.
Why?
They take it for granted that there is obviously something wrong here. Are users actually concerned about this? I think most people would not care. Is this just manufactured for the exposure? Is this just another example of OpenAI's condescending moral superiority complex?
The cases brought forth by Marvin Gaye's family [1] showed that some judges will declare copyright infringement even if the melody, harmony and rhythm are different. Note that the author saying he reverse-engineered the original song in question probably had something to do with it, so in the end intent and artistic perception will always remain factors that no computer function can compute.
[1] https://en.m.wikipedia.org/wiki/Pharrell_Williams_v._Bridgep...
https://openai.com/index/how-the-voices-for-chatgpt-were-cho...
Anyway, let’s say he was negatively affected by that relationship, IIRC.
Reminds me a bit of this: https://www.uniladtech.com/news/ai/man-married-hologram-no-l... , up to you if you find people developing strong feelings to inanimate objects that can’t care less, dystopic or not.
You do know the whole modus operandi of AI companies is "screw laws, take it all we'll see what happens" right ? That they actively lobby to not respect IP laws because they couldn't exist if they were enforced ? That they do not care about the negative impact they might have on anything ?
I'm no proponent of IP myself but I don't take from others to redistribute for my sole profit.
Seems to be you'd have to be pretty prejudiced against AI to say "yes".
I have no insight to draw from this, I'm just fascinated by it.
But I have tried many times. It does sound similar to me. Ask her to recite poetry with natural pauses and expression.
Yeah, for sure, I agree with all that. In this case, openai says they got a voice actor though so I'm not sure if IP applies here. Because of the movie SJ was in, I could honestly see a lawsuit going either way I guess
But I'd also argue that flirtatiousness is a very good match for the character of Samatha in "Her" -
I'm worried that in the future somebody might try to create their own voice to regain it as an accessibility tool after a medical event, and they're going to have their voice literally stolen from them under the fact that it might sound like another famous person 90% of the time.
In the music industry it's actually a known thing that if somebody famous sounds like you, you basically have no real ability to become a star or sell your music without problems. Sure it happens, but it's honestly the same problem many actors that look alike go through but just with sound.
This is pure censorship/money grab, and to make it worse, the voices sounded nothing alike.
This is false. The TTS conversation feature has been out for quite a while.
She did the voice for Her, a movie about an AI voice. If you check IMDB she's literally credited as "(voice)". Specifically in the realm of AI, her voice is her entire celebrity status.
2024-05-20T07:16:23.679Z
OpenAI to Pull Johansson Soundalike Sky's Voice From ChatGPT
OpenAI is working to pause the use of the Sky voice from an audible version of ChatGPT after users said that it sounded too much like actress Scarlett Johansson.
The company said that the voice, one of five available on ChatGPT, was from an actress and was not chosen to be an "imitation" of Johansson, according to a blog post.^1 Johansson played a fictional virtual assistant in the film Her, about a man who falls in love with an AI system.
The voices are part of OpenAI's updated GPT-4o, which debuted earlier this month and can reply to verbal questions from users with an audio response.
1. https://openai.com/index/how-the-voices-for-chatgpt-were-cho...
Maybe I have missed something. Is this really the fulltext of the "article"??
Every time I tried to suggest that maybe, LLMs and GAN tools don't make creativity easier but lazier and emptier or that this technology area is parasitic off human culture, every time an OpenAI junkie told me, "hey, perhaps humans aren't much different from LLMs", or someone said artists are derivative too and don't really deserve any more protections or are "gatekeeping art"...
... my anger at the time is vindicated every time these greedy, cynical wretches that the US tech industry has raced to anoint are taken down a peg because of their own very obvious greed and expedience.
I am loving this.
I may also be shallow in feeling a measure of glee that Microsoft is racing forward to shoehorn this utter toxicity into every corner of their product range, just in time for their customers to fully understand how it reeks of contempt for them.
They approached Johansson and she said no. They found another voice actor who sounds slightly similar and paid her instead.
The movie industry does this all the time.
Johansson is probably suing them so they're forced to remove the Sky voice while the lawsuit is happening.
Nothing here is shitty.
Like that "Rebel Moon" on Netflix was how to NOT do it, with tons of stupid exposition spelling out stupid details that didn't make any sense.
Versus "American Sniper" that was so evenly portraying all sides, that Right leaning people thought it was a liberal movie, and Left leaning people thought it was Right Wing propaganda. It was all so well done you could read into it a lot of your own feelings.
So "Her" was about the danger of technology. And at the end there were some scenes that you could read into how a lot of people were falling for this phone app and things were going downhill. But, it wasn't clear cut, the movie is really good at splitting the difference on how the app was also making people be happy, and was helping them.
At its core, Her was a beautifully-shot love story between two flawed beings, nothing more.
Also, they proceeded to ask her for rights just 2 days before they demoed the Sky voice. It would be pretty coincidental that they actually didn't use her voice for the training at all if they were still trying to get a sign off from her.
For every creative task I've given an LLM in the last 2 years, if I cared at all about the output, I ended up redoing it myself by hand. Even with the most granular of instructions, the output feels like a machine wrote it.
I have yet to meet anyone who felt any kind of emotion from generated art, except for "wow, it's cool that AI can make this". That's because (imo) art comes from experience, and experiencing is absolutely not what LLMs do.
Meanwhile, my dad, whose AI experience amounts to using MS Copilot "two or three times," is sending me articles about Devin, and how it's over for software engineers.
https://daringfireball.net/linked/2024/05/20/openai-johansso...
These rights should have their limits but also serve a very real purpose in that such people should have some protection from others pretending to be/sound like/etc them in porn, ads for objectionable products/organizations/etc, and all the above without compensation.
Have you ever observed how difficult it is to _remember_ AI generated pictures?
I can think of only one AI-generated art thing that has stuck with me, and it's because of the enormous amount of effort the guy using it went to generating really genuinely creepy fake photos to go with a plausible but fake story (about a lost expedition in the early era of photography).
I thought at the time, OK, maybe people will do creative things with it. Maybe I am wrong.
Except that months on I can't remember any specific detail of any of the photos in enough detail to visualise it. Only the emotion and the feel, which could have been evoked by that talented person entirely without Stable Diffusion.
There is something about AI generated photos, in particular, that confounds my ability to remember the image (as a photographer)
- they used Johannson's actual voice in training the text to speech model
or
- a court finds that they violated Johannson's likeness.
From hearing the demo videos, I don't think the voice sounded that similar to Johannson.
But hiring another actor to replicate someone you refused your offer is not illegal and is done all the time by hollywood.
I do like that many people have learned to recognize the writing style and visual aesthetic, and are rejecting it.
> maybe people will do creative things with it
_Some_ people will do _some_ creative things with it, but most people will use it as a shortcut—as long as there's some kind of output, they couldn't care less about the quality. How much of correspondence is just an LLM summarizing what a different LLM wrote? If the internet wasn't dead before, this is surely killing it.
Everything about OpenAI speaks of people who do not put great value on shared human connections, no?
Hey, I like that artist. I am going to train a computer to produce nearly identical work as if by them so I can have as many as I like, to meet my own wishes.
Why is it surprising that it didn't really cross their mind that a virtual girlfriend is not a good look?
This is not an organisation that has the feelings of people central to its mission. It's almost definitionally the opposite.
This is the thing that gives me hope -- inquisitive people who have no idea how ChatGPT does what it does can point out ChatGPT-generated text. It's more difficult with GAN-generated images but in the creative community I am part of, some people are very literate about this already.
I mean…if you check the IMDB page you’ll see she’s done a lot more than that…and is known widely for her role as Black Widow in the Avengers movie series.
So, I think she’s got more to her status than “the voice”.
I’m sure you were trying to say in this OpenAI voice rights context, but I think it’s important to consider she is indeed a bonafide not-just-a-voice A-list Hollywood Celebrity.
I also think it is tipping their hand a bit. I know companies can do multiple things at once, but what might this flirty assistant focus suggest about how AGI is coming along?
No matter how much you filter and purify it, puke will rank.
Let someone else swim against the tide if they feel like that's the best use of their time.
From technical standpoint, a finetuned voice model can be built from just few minutes of data and GPU time on top of an existing voice model, almost like how artists LoRAs are built for images. So it is entirely within possibility that that had happened.
Funny thing is, when an attractive female goes against a corp... the simps picked the corp... we are well off into a Blade Runner 2 digital girlfriend scenario aren't we?
If any enterprising soul wants to join up and start selling chatGPT enabled anime body pillows, DM me, I have a supplier ready.
This strongly suggests they weren't trying to get her voice until the last minute (would have been too late for the launch) but, rather, they had already used the other actress, and realized they were exposing themselves to a lawsuit due to how similar they were.
It was a CYA move, it failed, and now their ass is uncovered.
Johansson was foolish to turn this down. This all sounds like she realized the mistake, regretted it, then sent her legal team to pursue this frivolous cease and desist out of spite.
I'm disappointed that OpenAI didn't see this for what it is, and decided to comply instead.
But quite a lot of people understand the difference, at a visceral level, between a painting made by an individual amateur artist and a painting made for selling at one of those Fine Art chains, or the difference between something rough and charming and a painting you might have seen in the 90s while trying to locate the loo in a UK branch of McDonalds.
People's instinctive artistic "literacy" is often surprising.
Making a shitton of money.
Sorry.
Probably this could indeed make them "win" (or not lose rather) in a legal battle/courts.
But doing so will easily make them lose in the PR/public sense, as it's a shitty thing to do to another person, and hopefully not everyone is completely emotionless.
The effect of the DMCA on youtube has shown that IP seems to apply even if something vaguely looks or sounds like the original. "Apply" not meaning "is true", but "has consequences".
Neat
OpenAI clearly wants the “her” association. It’s already the favourite voice and projects many of the existing and imagined capabilities really well.
So they tried to pay for it, but it was either too expensive or blocked by SJ.
Her/teams comments read like a negotiation.
But I suspect they’ve overplayed their hand. At best they may get a settlement and OpenAI will modify the voice to make it sound less like SJ.
But parent shouldn't feel too proud of their prognostication skills. OpenAI is a venture of Sam Altman and Elon Musk, so how could it be anything other than what it is? You'd have to be insanely naive about SV (and, more broadly, what "non profits" of billionaires in any sector even are) to assume this was ever born of altruism.
Yet the vast, vast majority did and a still large proprtion continue to proclaim that these projects were born of altruism and continue to serve these altruistic goals and these people are most incredible altruistic humans to ever grace this fine planet of ours.
Like, lets be real here. This wouldn't be the first time they would be using material without the right to them and I don't expect this to change any time soon without a major overhaul of EVERYTHING IN THE COMPANY and even then it will probably only happen after lawsuits and fines.
I also don't profess surprise at who OpenAI have turned out to be. Rather I am surprised that other people are surprised.
It's not a heel turn, except in their wider cultural fortunes. It has been obvious to me from literally day one that everything to do with DALL-E and ChatGPT and onwards is bad for culture. There has never been anything other than creepy, dystopian, Black Mirror overtones.
But the valley falls for hucksters every time. And it's often the same hucksters.
It's not just schadenfreude (which I admit is unattractive, if beguiling.)
It also gives me hope that ordinary people are beginning to get to grips with the idea that they don't have to accept or be excited for new technologies just because they are new technologies, and that the people bringing new technologies don't have to be good people just because they are capable people. Seemingly smart people can be intellectually and morally lazy.
I have no obligation as a techie person to be excited about AI, or to be default-positive about the "leading firm", or to give the benefit of the doubt, or anything like that. There's no moral rule that one should be positive about new technology until it's proved bad. This is a classic tech industry false belief.
OK so the fall is not happening as quickly as Juicero. But it's a start.
What's your case for why should I not be happy?
Even though it threatens their livelihoods and is parasitic off their work?
It's not disheartening at all: it's positive.
> Johansson was foolish to turn this down. This all sounds like she realized the mistake, regretted it, then sent her legal team to pursue this frivolous cease and desist out of spite.
This all sounds like absurd wishful thinking.
If an actor is saying no and you have a certain creative vision then what do you do?
Johansson doesn't own the idea of a "flirty female AI voice".
It's much closer to RJ than SJ... MUCH much closer. Enough that if it's not her, I'll be genuinely surprised.
Explain why you call a famous actress foolish if she refuses to give a corporation permission to use her voice. Your entire argument is built on this opinion.
There are other AI companions as well in the film.
It reads like projection of some other feelings or unstated interests.
I have no science to back this up, mind you. But I struggle to recall details of these images (I also believe I have a limited form of aphantasia so it could just be my flawend noggin)
But I will take your point ;-)
That's exactly what was done when Jeffrey Weissman replaced Crispin Glover in Back to the Future Part II.
Answer is: it’s not. To all three. And collectively we can decide to be better. This is why artists are pushing back. One day perhaps the tech world will understand that they aren’t Luddites but instead champions for humanity.
Really? AI has lots of potential but so far the big uses of the recent title wave have been an enormous increase in the creation of visual and text-based sludge, barely usable for anything serious most of it, by hustling online marketers and social media spammers.
Even where tools like GPT are used productively by people to simplify their business processes and so forth, every piece of information they claim has to be scrutinized for hallucinations to the point of them being useless as much more than idea generators for contexts where factual correctness isn't important...
Yay!
> Rather than write George out of the film, Zemeckis used previously filmed footage of Glover from the first film as well as new footage of actor Jeffrey Weissman, who wore prosthetics including a false chin, nose, and cheekbones to resemble Glover. [...]
> Unhappy with this, Glover filed a lawsuit against the producers of the film on the grounds that they neither owned his likeness nor had permission to use it. As a result of the suit, there are now clauses in the Screen Actors Guild collective bargaining agreements stating that producers and actors are not allowed to use such methods to reproduce the likeness of other actors.[
> Glover's legal action, while resolved outside of the courts, has been considered as a key case in personality rights for actors with increasing use of improved special effects and digital techniques, in which actors may have agreed to appear in one part of a production but have their likenesses be used in another without their agreement.
https://en.wikipedia.org/wiki/Back_to_the_Future_Part_II#Rep...