It's a neat demo. I don't know how I feed about cloning the voice of a living person, even a 'public' figure like a presenter. I wouldn't be surprised if it eventually becomes illegal, although it'll likely be as difficult to enforce as normal casual copyright infringement.
I expect the existing laws will be extended. I had a colleague tell me about looking into how much Morgan freeman wanted for a 15s commercial ($$$$$$). The alternative he used was a “sound alike” voice actor, for which Morgan Freeman received royalties to the tune of $$$$$.
Why would Morgan Freeman be entitled to royalties for a commercial where the voice over sounded similar to his voice... unless he actually sold the rights to represent him with a vocal likeness in the commercial (AKA he didn't do the voice over, but the commercial was able to claim the voice over was by Morgan Freemen, or that it represented Morgan Freeman's opinions... like when a book by a celebrity is actually written by someone else)
You can't use a Tom Waits (or Morgan Freeman) sound-a-like because you're profiting off their work and likeness. Unless the celeb sells a licensed sound-a-like.
What if my voice sounded like Tom Waits or Morgan Freeman and I did commercials? It doesn't and I don't I'm just curious how this makes sense. Is it just if the voice is advertised to sound like him?
> The agency came back to Frito-Lay with a parody song inspired by Tom Wait’s ‘Step Right Up’, featuring the same Carni hollerings as the Small Change track
If you did straight-up parodies of their songs, yes.
Sounding like someone isn't just having the same timbre of voice, it's cadence, accent, diction (both definitions). I'm skeptical how much two unrelated people could just happen to sound alike, especially if one of them is famous for their voice. Compare Larry David's natural speaking voice to how he speaks when impersonating Bernie Sanders. Those are two men from the same time and place and background, and there's still quite a gulf.
Now perhaps you could just happen to sound like a famous actor who isn't well known for their voice, like John krasinski or something, who does voice over ads but isn't super distinct. That would probably be fine but you're still going to have to slog through the world of voice talent auditions
I don't think with the current copyright system you would owe another artist money when recording a commercial just because your voice happens to sound like theirs. But with AI perhaps it's "fair" (given the assumption that there's any fairness to copyright in the first place) to expect a creator to provide some kind of compensation to the source material that their AI was trained on.
Case in point, this bot was clearly trained on David Attenborough samples, perhaps it can't be used commercially without purchasing some rights to the man's likeness.
> I don't think with the current copyright system you would owe another artist money when recording a commercial just because your voice happens to sound like theirs
No, its a right of publicity/right of personality issue, not a copyright issue. While that's not a federal law issue in the US, its an issue in some US jurisdictions, and its a national law issue in a number of non-US jurisdictions.
The mistake is thinking “would not owe money under the current copyright regime” means “would not owe money”: there are more laws in the world than just copyright law.
What would happen if a voic was trained on Morgan Freeman, Jack Nicholson, Robert De Niro, George Clooney and tuned to not sound like any one individual but have the characteristics we find appealing. Could that be used for its own copyright claim?
I believe the answer is almost certainly yes under current law.
Whenever the topic of copying material for model training purposes arises on hn, however, a number of folks seem convinced such use falls under the rubric of “transformative” and will be considered fair use.
Do you just believe this or was this tested in court yet? I haven't been following the various lawsuits on this issue closely but I had thought that, so far, they were leaning towards fair use.
I believe it, and I don’t think it’s been significantly tested.
It’s hard to see how mixing (copying!) a bunch of books into a computer algorithm to produce (commercially!) content that might itself easily be incorporated into a book in the same market could possibly be fair use.
There is a significant difference between "stealing" artistic ideas, and impersonating individual humans.
How folks can think it's reasonable to impersonate Morgan Freeman and use his voice to sell, for example, crypto crap. It's gross and should absolutely be illegal.
In that case though (and the Bette Milder precedent), the sound-a-like was used with a song that the artist was associated with, which implied that it was the actual artist, or at least deliberately invoked their likeness in mind of the viewer.
I'm skeptical that using a narrator that happened to sound like David Attenborough or Morgan Freeman would be enough by itself to qualify for right of publicity unless it included obvious references to prior work, catch phases, or the like.
Difficult to find a Morgan Freeman sound-a-like if they're not marketing themselves as a Morgan Freeman sound-a-like. And if they are (and obtaining more customers as a result), it's easier to pay Morgan Freeman royalties than come up with persuasive arguments that selling 'voiceovers that sound like Morgan Freeman' isn't profiting from deliberately invoking his likeness...
> Difficult to find a Morgan Freeman sound-a-like if they're not marketing themselves as a Morgan Freeman sound-a-like.
How about this scenario: they publish a youtube video, then someone says in the comments: hey you sound just like morgan freeman, then the company looking for the voice performance googles for morgan freeman, finds the comment and contacts them.
So then in court the google for Morgan Freeman and the comment you sound like Morgan Freeman come out and then you owe Morgan Freeman $$$$$$$$$ because he had to go to court over it.
Think if I wanted to earn a living as a voiceover artist who happened to sound like a very famous actor I'd rather advertise my USP and pay royalties on the much higher earnings than only pick up business from companies that make hiring decisions based on random chance or YouTube comments ...
If your voice is the twin of a celebrity's, but you don't benefit from the likeness (for example, not being paid for voice acting where your identities could be confused), you'd probably have a strong defense if the celebrity tries to extort royalties from you.
The rest is shades of grey. For example, if you're an incredible voice imitator and plan to do shows, negotiations would be wise. Even Weird Al asked for permission when he didn't had to, legally speaking.
Weird Al's latest movie is a mess, but so surreal I keep remembering it months after watching it. It as if the UHF movie carved a niche in my brain for his weird movies.
An easy workaround for this is to dress your German speaker in a Nazi uniform, now suddenly nobody wants to. be associated with them, even if you keep the unkempt white hair.
If the voice was chosen specifically to sound like Morgan Freeman, for the purpose of making listeners think it's Morgan Freeman when they hear it, it seems reasonable for it to run afoul of Morgan Freeman's right to publicity.
I owe nothing to Morgan Freeman just because I sound like him, unless I'm intentionally misleading listeners by advertising his name or implying as much
If they can reasonably argue that you're being hired because you sound like Morgan Freeman and you wouldn't get the job otherwise, then they likely have a case.
Not if "sound like Morgan Freeman" is a proxy for "have a nice-sounding voice". They could prefer people who sound like Morgan Freeman even if they are not representing to listeners that the voice is that of Morgan Freeman. Big distinction there.
Is it a case of (a) they like Morgan Freeman's voice as a general preference (and therefore want someone who has a nice-sounding voice such as Morgan Freeman's for features such as how it sounds smart, deep, engaging, I don't know) or (b) they want someone who sound like Morgan Freeman to somehow represent to listeners that it's Morgan Freeman's voice?
Gathering, retaining, using individual voice recordings for business purposes may be regulated, though. Already is in places, even if enforcement and precedence is lacking.
So there'd be no legal way to produce or share the model generating such a voice or its output.
My father had a friend who was a well recognised commercial voice over artist , and told me about his fees. There was a fee for doing the actual work, but a much bigger exclusivity/exposure fee designed to prevent his voice getting over used to the point where future work dried up.
The extremely obvious legal problem is, ALL simulated voices sound like some real person somewhere, it's just a matter of finding that person.
I don't see a solution to this, so I'll tend to believe the cynic's viewpoint that the final written law will be in favor of whoever has the most lawyers.
It's actually simple. Take David Attenborough's name out of the picture.
"A generic British guy is now narrating my life" isn't as funny as implicating a known celebrity, which is where the legal issues would arise. A generic British guy isn't going to sue you.
Hell, you might even be able to get away with "A generic British guy who sounds like David Attenborough is now narrating my life."
Sure. The difference is some voices, like David Attenborough or Samuel Jackson are instantly recognizable to a large audience while someone who isn't famous' voice is not.
Where to put the bar is a specific issue, but the exemplars in their category stand out as recognizable.
>ALL simulated voices sound like some real person somewhere
By the same logic, all simulated faces look like some real person somewhere, but I wouldn't expect to simulate a face that looks like Brad Pitt and then use it in a TV commercial
While I agree, I also did thoroughly enjoy a recent Youtube channel that did Warhammer 40K lore in Attenborough's voice. It recently got taken down for obvious reasons.
Makes you wonder if in a few years there will be licenses available for using someone's voice for such endeavours.
I have been creating audiobooks from ebooks for a while now. I just grab an .epub and turn it into a pretty good audiobook in minutes for completely free using local models.
Personally, I don't really see a problem with it as long as it's not used deceptively. In this case, it's pretty clear that it's not really David Attenborough speaking.
Outside of that, I am not sure I'm seeing a moral or economic argument for making voice copyrightable. The usual argument about incentivizing creation doesn't apply here.
That said, it'll be interesting how the legal side of this shakes out. Does one own their likeness and voice? Who decides what that is? There are plenty of voice actors that can do a great Morgan Freedman or trump. How is that different?
Cloning the voice of a dead person, to me, feels even icky-er.
Indigenous Australians would probably feel the same, too, considering they avoid contact with pictures or recordings of deceased people.
Shouldn't be illegal, but I could understand it if we see laws that require explicit disclosure when voices of living people are simulated. At first glance that seems like the right compromise.
I just did this. I couldn't resist. I looked into the camera and Werner said:
> In the quiet chaos of an ordinary room, a lone figure peers into the abyss of a lens, a silent sentinel in the company of walls adorned with fragments of a life well-lived. The ambient light casts a celestial glow atop their crown, an unintentional halo in a moment of serendipity. The gaze, piercing yet calm, confronts the onlooker, challenging the nature of existence through the simple act of capturing a fleeting second in the digital ether. Here stands a human, enshrined in the mundane yet extraordinary tableau of their own making.
“… and here we have the common Twitter poster. Their attempt to gain favour with their peers and, therefore, popularity with potential mates falls foul once again.”
Given the Attenborough Lore channel for Warhammer received a cease and desist, you can bet that his legal team will be busy suing the shit out of people like this.
Same. When I logged into YouTube and saw "Scholar's Lore", my heart sunk and I immediately knew what had come to pass. Attenborough is fiercely protective of his voice.
Was it noncommercial and done in the spirit of satire (surrealism?) - in which case surely they'd get a pass - but I guess they just didn't want to go bankrupt trying to defend themselves?
It's a shame then - when you compare it to the situation between the nerdcore gangsta-rapper alter-ego of Stephen Hawking: MC Hawking; who was created without authorization by Dr. Hawking, and sold real physical CDs for a profit (and I own one!), but who ultimately received an endorsement from the late physicist who took it all in good humour - and arguably Stephen Hawking is a bigger public figure than David Attenborough.
...if it really does comes down to just the personal approval by the subject themselves (and not their legal/marketing/legal teams acting without them even being aware) then I'm concerned what it means for artistic liberty vs public figures.
I don't believe it was intended to be satirical. The depth of worldbuilding and lore in Warhammer affords an immense opportunity for Games Workshop to actually do some kind of series like what was created with Attenborough's likeness. His voice is his main source of income for his estate nowadays, so it makes sense he doesn't want it replicated, especially without any permission being granted.
Too bad they'd never do that and would probably do a worse job if they did. There should be carve outs for fans to make things that bloated corporations will never have the ingenuity to come up with or the balls to carry out.
That was such a great channel. The voice was paired with pitch perfect writing as well. The script had exactly the sort of flow that an Attenborough BBC program would have, plus it was actually really informative.
He really should re-dub the entire series with a unique AI generated voice and repost. They wouldn't be as good, but still pretty close.
He is doing exactly that! It's not as unique as Attenborough's voice, but it's nice to see every video being reuploaded slowly. He's even taken feedback from the previous videos into consideration to "remaster" them.
Why do you think it is gross? Also, what does gross mean in this context? It's a very ambiguous term. Would it be less gross if this was a meme video where someone does an impression of Attenborough and just narrates what the person being filmed is doing in a comical way, there are a million of those probably.
A pitch perfect Attenborough impression by a human would be an expression of talent and itself a kind of artwork. We understand intuitively that perfect impressions of famous people are scarce. When we watch something like this, we see that those impressions will be cheap, instantaneous and universally available; there is only the negative aspect, and none of the wonder at what it took to pull off.
Perhaps for a few minutes we're going to be marveling at the achievements of developers who make these things! But most people here, and, in the near future, everybody else, understand that a programming accomplishment today is a boxed up SKU next year. Most likely, in a very short period of time, the overwhelming majority of people making use of this technology will have had no hand in the accomplishment of building it.
This is a good thing. It democratizes access to a product that very, very few could afford when it required a lot of human talent and time to produce. In my experience, I only marvel at impressions done as an unexpected party trick and the novelty quickly wears off. If I hear a voiceover that sounds very much like some famous person I'm not imbued with wonder I just perceive that the audio was well produced and don't give it a second thought. I think this kind of automation may even be very good for people with unique voices who are looking to get into voice acting, they should be able to pay some service to clone their voice and then offer it on demand for a small fee. The amount of work they would have to put into this is greatly reduced compared to if they had to do all of the voice acting themselves and they are able to take more customers since the work can be done in parallel. It's like digital publishing for voice acting with the same benefits and detriments e.g. your voice can be cloned like your ebook or song can be torrented but you also get a much bigger market and it's much easier to enter it.
Interesting ... and awful ... it sharpens how I feel about it knowing that the man is nearing the end of his life. This is absolutely not how I want to experience and remember the man.
This is astounding. This is the kind of demo that we could only dream of just five years ago, now some hacker dude on Twitter can just throw this kind of thing together. Seriously cool.
It’s exciting to imagine that we can soon enter the age where people will be able to voice their opinions in their own voice and choice of words, long after they are dead and gone from this physical world. Perhaps the first version of some digital afterlife.
I too long to retain some corporeal presence as a box, filled with an llm trained on my life's collected texts and utterances. A featureless grey cube with only two buttons marked "more snark" and "more grossness", to be dusted off and interacted at every christmas by my great-grandchildren.
But like, you can't know what someone would have said in the future if they aren't there to say it. No amount of LLM improvements would be able to know the real internal thoughts and memories of someone. You could guess, sure, but you don't know with any certainty.
People change. Small things change people - seeing a car speed by might alter your opinion on some variety of things, but no LLM was there to capture that moment.
It will always be a guess. You can never know for sure what someone who is deceased would think about something in the future since they are simply not there.
And of course anyone can use your voice ti voice any opinion. Want Ghandi saying his words are backed with nuclear weapons? Want to portray Hitler as a nice loving chap? And that’s before you get onto current politicians - https://www.theguardian.com/politics/2023/nov/10/faked-audio...
I sincerely find this to fall into the proscriptions against necromancy present in most ancient belief systems. The dead should not be made to speak the words of the living. And we should not create for ourselves any illusions about the completeness or finality of death.
It's not so much that it's a moral transgression as that it will undermine and corrupt our own understanding of what it means to be a living person.
Turns out, it doesn't mean much to be a living person. Ideas are what matter to society, not the individual that spawns them. When you kill a revolutionaire, you are only killing a man.
Depends on your framing I guess. I live life as a person not as a society, and so being a person is quite meaningful to me. I don't know what it would be like to be an entity that experiences it otherwise and I'm glad for that.
This will be a fairly trivial problem to solve. Each spoken word can be cryptographically signed on some kind of distributed public ledger and unless the words originate from a verified origin you cannot assume something you hear was indeed spoken by the original source LLM.
Only verified sentences coming from your LLM clone can be considered the actual you.
I am starting to believe a lot of people working in ai are sociopaths. Their fixation with impersonating people and stealing ip points firmly in that direction.
I have really enjoyed the new show Scavengers Reign. It does an amazing job of showing an alien planet's biodiversity. I have fantasized about a cut of just the nature scenes with Attenborough narrating. It seems like this make an attempt at that.
YouTube is developing revenue sharing for AI-generated music [1]. I wonder if they will extend revenue sharing to non-musical content like the narration in OP.
Pretty cool, a couple of months back a friend asked me if I could get him a Carl Rogers audiobook, I forget which one. I told him sure and jokingly asked if he wanted David Attenborough reading it. He said yes. I had myself a challenge. Long story short, I was able to create an audio book for a friend, the voice was meh!
I think you shouldn't be able to distribute code / model like this without permission...
...but...
What if you distributed code that when I ran it, and plugged in each of the Blue Planet / Planet Earth DVDs, it would learn how to do this, with a local model?
Wonder how he’s deciding which frame to feed to judge.py, given it would be tremendously wasteful to call gpt4vision on every frame. Maybe there’s some logic to detect meaningful drifts?
I think this is nasty to provide as a service to others because the voice didn't consent. I don't agree with others when also critiquing personal use. Personal use should be fair use as long as you aren't stealing or mining a service to minimize some personal cost.
So is this a text to speech model while mimicking somebody's voice? That is pretty cool.
One thing I don't understand is how you get specific models for someone's voice. Do you fine tune to that voice?
The issue of style transfer is not something I understand for audio. Anyone have example papers to read? I am familiar with MuLan, MusicLM, and MusicGen.
In the tweet he says he is using https://elevenlabs.io for the voice, which is currently the best quality text-to-speech service and allows for voice cloning too.
And Elevenlabs is saying this about cloning other people's voices:
Can I clone anybody's voice?
Yes, as long as you have their consent. For Professional Voice Cloning, we have integrated robust security measures to make sure you can only clone your own voice. Unless you share it, your voice belongs and is available only to you.
But it looks like it's not enforced if the example in the tweet is possible.
But even if it would be enforced by the company, open-source models are getting close to them and they are harder to control. The public or commercial use of the output from these models can (and probably should) be limited by laws, but if the users want to use it privately (for example to narrate themselves) there is not much you can do to prevent it especially with the progress in this tech it will become more trivial to do it for anybody.
It actually is enforced. They have a verification system where they show you random text on the screen and you have to read that into the microphone. So I don’t know how the tweet author got around that.
