zlacker

[parent] [thread] 14 comments
1. worsts+(OP)[view] [source] 2024-05-20 23:37:58
Most of the reactions here are in unison, so there's little left to contribute in agreement.

I'll ask the devil's advocate / contrarian question: How big a slice of the human voice space does Scarlett lay a claim to?

The evidence would be in her favor in a civil court case. OTOH, a less famous woman's claim that any given synthesized voice sounds like hers would probably fail.

Contrast this with copyrighted fiction. That space is dimensionally much bigger. If you're not deliberately trying to copy some work, it's very unlikely that you'll get in trouble accidentally.

The closest comparison is the Marvin Gaye estate's case. Arguably, the estate laid claim to a large fraction of what is otherwise a dimensionally large space. https://en.wikipedia.org/wiki/Pharrell_Williams_v._Bridgepor...

replies(4): >>jakela+25 >>teloto+58 >>TaroEl+2K >>wrapti+zO
2. jakela+25[view] [source] 2024-05-21 00:08:00
>>worsts+(OP)
(Not a lawyer) I think the issue is not just that they sound similar, but that OpenAI sought to profit from that perceived similarity. It’s pretty clear from Sam Altman’s “her” tweet that OpenAI, at least, considers it a fairly narrow slice of the human voice space.
replies(1): >>worsts+Ia
3. teloto+58[view] [source] 2024-05-21 00:27:10
>>worsts+(OP)
This is almost an identical case, and resulted in a ruling favorable to Midler, whose voice was imitated in that case: https://en.m.wikipedia.org/wiki/Midler_v._Ford_Motor_Co
replies(1): >>worsts+3a
◧◩
4. worsts+3a[view] [source] [discussion] 2024-05-21 00:39:08
>>teloto+58
It's indeed pretty similar. However, it involved singing a song Midler was known for. This case is at best peripheral to the movie Her, in that the OpenAI voice does not recite lines from the movie.
replies(1): >>pseuda+Wz
◧◩
5. worsts+Ia[view] [source] [discussion] 2024-05-21 00:43:39
>>jakela+25
That's true. As I suggested, this case may well be open and shut. But what if it was Google's or Meta's voice that sounded exactly like Sky, i.e. without a history of failed negotiations? The amount of "likeness" would technically be identical.

Are companies better off not even trying to negotiate to begin with?

replies(2): >>jakela+jc >>MrMetl+er
◧◩◪
6. jakela+jc[view] [source] [discussion] 2024-05-21 00:53:47
>>worsts+Ia
Not sure legally. I chose the words “perceived similarity” intentionally to encompass scenarios in which the similarity is coincidental but widely recognized. Even in that case, I believe the original person should be entitled to a say.
◧◩◪
7. MrMetl+er[view] [source] [discussion] 2024-05-21 03:13:32
>>worsts+Ia
It'd be fine to not negotiate if you weren't going to use a voice that sounded famous. If I were offering something that let you make any kind of voice you want, I would definitely not market any voice that sounded familiar in any way. Let the users do that (which would happen almost immediately after launch). I would use a generic employee in the example, or the CEO, or I'd go get the most famous person I could afford that would play ball. I would then make sure the marketing materials showed the person I was cloning and demonstrated just how awesome my tool was at getting a voice match.

What I wouldn't do is use anything that remotely sounds famous. And I would definitely not use someone that said "no thanks" beforehand. And I would under no circumstances send emails or messages suggesting staff create a voice that sounds like someone famous. Then, and only then, would I feel safe in marketing a fake voice.

replies(1): >>worsts+zs
◧◩◪◨
8. worsts+zs[view] [source] [discussion] 2024-05-21 03:25:07
>>MrMetl+er
Sounds judicious. You probably wouldn't get sued, and would prevail if sued. However, the question of how much human voice space Scarlett can lay claim to remains unsettled. Your example suggests that it might be quite a bit, if law and precedent causes people to take the CYA route.

Consider the hypothetical: EvilAI, Inc. would secretly like to piggyback on the success of Her. They hire Nancy Schmo for their training samples. Nancy just happens to sound mostly like Scarlett.

No previous negotiations, no evidence of intentions. Just a "coincidental" voice doppelganger.

Does Scarlett own her own voice more than Nancy owns hers?

Put another way: if you happen to look like Elvis, you're not impersonating him unless you also wear a wig and jumpsuit. And the human look-space is arguably much bigger than the voice-space.

replies(2): >>jakela+Jt >>kelnos+VC
◧◩◪◨⬒
9. jakela+Jt[view] [source] [discussion] 2024-05-21 03:38:52
>>worsts+zs
I know toying with these edge cases is the “curious” part of HN discussions, but I can’t help but think of this xkcd: https://xkcd.com/1494/
replies(1): >>worsts+mv
◧◩◪◨⬒⬓
10. worsts+mv[view] [source] [discussion] 2024-05-21 03:54:24
>>jakela+Jt
HN discussions, grad school case studies, and Supreme Court cases alike. Bad cases make bad laws, edge cases make extensive appeals.
◧◩◪
11. pseuda+Wz[view] [source] [discussion] 2024-05-21 04:41:18
>>worsts+3a
The decision said Midler's voice was distinctive. Not the song.
◧◩◪◨⬒
12. kelnos+VC[view] [source] [discussion] 2024-05-21 05:12:26
>>worsts+zs
> However, the question of how much human voice space Scarlett can lay claim to remains unsettled

I don't think it's that unsettled, at least not legally. There seems to be precedent for this sort of thing (cf. cases involving Bette Midler or Tom Waits).

I think the hypothetical you create is more or less the same situation as what we have now. The difference is that there maybe isn't a paper trail for Johansson to use in a suit against EvilAI, whereas she'd have OpenAI dead to rights, given their communication history and Altman's moronic "Her" tweet.

> Does Scarlett own her own voice more than Nancy owns hers?

Legally, yes, I believe she does.

replies(1): >>MrMetl+OA2
13. TaroEl+2K[view] [source] 2024-05-21 06:30:42
>>worsts+(OP)
My concern is that cases like this would set the precedent that synthetic voices can't be too close to the voice of a real, famous person. But where does that leave us? There's been lots of famous people since the recording age, and the number is only going to increase. It seems unlikely that you can distinguish your fake voice from every somewhat public/famous real voice in existence, especially going forward. Will this not result in a situation where the synthetic voices must either sound clearly fake and non-human to not be confused with an existing famous voice, or the companies/producers must in every case pay royalties to the owner to a famous voice that sounds similarly close, even if their intent wasn't even to copy said voice or any that are similar, to avoid them getting sued afterwards? Are we going to pay famous people for being famous?
14. wrapti+zO[view] [source] 2024-05-21 07:14:32
>>worsts+(OP)
This sort of copyright seems completely unethical to me.

We have 8 billion people, probability of unique voice and intonation is extremely unlikely. Imagine someone else owning your voice. Someone much richer and more powerful. No entertainment is worth putting fellow human beings through such discrimination and cruelty.

◧◩◪◨⬒⬓
15. MrMetl+OA2[view] [source] [discussion] 2024-05-21 18:23:34
>>kelnos+VC
There are other ways public figures are treated differently in the courts in the US. It's much more difficult for them to prove libel or slander, for instance. They have to prove actual malice and intent, whereas a private citizen just has to prove negligence. I imagine "owning" their likeness at a broader sense is the flip side of that coin.
[go to top]