zlacker

What are the unique aspects of a sound? A lot of people look and sound stunningly alike.

As a recent example Baldur’s Gate 3, Andrew Wincott voiced Raphael, an npc-antagonist, who to my untrained ear sounded exactly like Charles Dance, and the character model had more than a passing semblance to Mr. Dance as well.

It was not a Charles Dance carbon copy but all aspects of the character were strongly aligned with him.

I’m wondering where is the line in style and personal aspects of one’s craft drawn.

Some of this is probably part of personal perception.

>>fsloth+(OP)
Wincott and Dance and are both British actors that began their careers on stage, so they have similar accents, cadences, and vocal mannerisms common to stage actors. For example, both of them speak like Patrick Stewart, another English who also began his career on stage. But otherwise they all clearly have very different voices: they have different timbres, vocal fry, and only one of them (Dance) can sing well and he has a surprisingly large vocal range (see his performance as the Phantom in Phantom of the Opera).

In this case, the actress selected for OpenAI was clearly selected for similarity to SJ. And that by itself would have been okay, because the actress is speaking in her natural voice, and SJ doesn't have a monopoly on voice acting...but OpenAI went further, and had the unknown[1] actress base her inflections, cadence, and mannerisms on SJ's performance in the movie Her. And Altman even tweeted the movie's name to advertise the connection.

The problem is that there is a well-settled case law stretching back over several decades that makes this a slam-dunk case for SJ, because it doesn't matter that OpenAI didn't "steal" her voice, they stole her likeness.[2] It wasn't just some unknown actress speaking in her own voice, it was an actress with a voice similar to SJ given lines and directing by OpenAI with the clear intent of mimicking SJ's voice performance in one of her more-famous roles.

[1] There is a very short list of a few actresses who both sound like SJ and do voice-over work circulating around Hollywood, so a lot of people have a pretty good idea of who it is, but nobody will identify the actress unless she identifies herself, out of solidarity.

[2] Likeness rights are quite strong in the U.S. They're even stronger in Europe.

>>gamblo+n4
Every single thing in your second paragraph is directly contradicted by the article at hand, yet you say them like they are established facts as opposed to things you just made up.

>>fsloth+(OP)
Who's Charles Dance? :)

As for Scarlett Johansson, I remember her from the Ghost in the Shell the live action movie controversy. Not fondly.

>>fsloth+(OP)
Try loading random voice file done by real (voice)actors into Audacity, switch view to spectrogram mode, drag down to expand, and compare it to yours. Professionally done voice should look like neatly arranged salmon slices, yours will look like PCIe eye diagrams.

You can also obviously compare multiple voice files recorded with similar sounding but different individuals, they rarely look similar on spectrograms.

>>numpad+67
Sure, except literally no one actually does this. You listen to a voice and it sounds similar in your head? That's who you picture when you hear it. Unless you're a robot I guess.

>>gambit+1a
The point is that human voices are technically and verifiably unique, tangential or perhaps antithetical to your/average person's perception.

>>numpad+Xc
I don't see how that's relevant here - court cases about situations like these are decided on the criteria of "if you show this to an average person on the street would they be able to tell the difference" not "if you load this up in a specialized piece of software and look at the spectrograph is there a difference".

>>gambit+4f
I think that will be a very clever and useful defense against CCTV footage and DNA analysis reports! Best legal advice ever.

>>numpad+ng
I don't understand why you are being sarcastic right now. Trademark cases are always decided on "if a person was shown this logo/song/whatever could they mistake it for the trademarked property of another company", not "well if you load it up in Paint you can see that some pixels around the edges are different so it's technically not the same logo your honour!".

>>gambit+Lh
so... your position is now in favor of SJ? I don't see consistency in your comments other than that the aim is to downplay uniqueness of voice to justify OAI's actions after the fact.

>>numpad+oi
No, my position hasn't changed - the average person on the street might think this voice sounds like SJ, but since SJ doesn't own exclusive rights to anyone else in the world sounding like her I don't think she has a legal ground to stand on, unless OpenAI pretended it is actually her. But I know for certain that the case will not be decided on spectographs of the voice.

>>ars+f6
No, I read the article.

WaPo's reporting states that the individual in charge of the interaction, Jang, modeled it after Hollywood movies, and worked with a film director specifically to accomplish this goal. And the executive responsible for the artistic decisions, CTO Murati, was conveniently not made available for WaPo to interview.

OpenAI has no credibility here, given its extensive history of dissembling as a company. If Her and SJ weren't the driving inspiration for the Sky voice, they would have made Murati available to explicitly refute those claims. Her absence speaks volumes.

And OpenAI dropping Sky immediately speaks even louder. It means that somewhere there is a smoking gun that would destroy them in court. [Edit: it turns out the smoking guns were already public: in addition to the CEO's Her tweet, his co-founder Karpathy explicitly linked the voice product to SJ. Game. Set. Match.]