Anonymous #1 3w

That would be against the law. Scarlet Johansson already made an example out of chatGPT for that exact situation, they wouldn’t want a class action suit on their hands. Also, they’d have nothing to gain from it in general, nobody goes on Duolingo looking for an expansive selection of voices to listen to, so why waste resources training extra voice models? It’s much more likely that they’re using transcript data in some capacity, if anything. Definitely not your voice though.

1

Anonymous OP 3w

It’s not illegal if they get your permission. Check the TOS and privacy policy.

3

Anonymous #1 3w

That is absolutely not going to hold up in court, lol. You need a lot more that a few lines in terms and conditions to get the right to use somebody’s likeness. They’re not hosting the model, so the interactions need to be forwarded to a company that does. It would be nice if they used a more private service like Proton though.

-1

Anonymous OP 3w

Ok but they’re literally telling you that they’re doing it so good luck I guess? Enjoy living in denial

5

Anonymous #1 3w

They don’t say they’re making copies of people’s voices, they say they’re using data to train models, the only models they utilize are conversational models. Basically they’re looking at how people respond so that they can train models that will have more flexible conversations, find parts of language learning that people struggle with, etc. There are a lot of shitty companies out there, but Duolingo is still alright (for now at least)

-1

Anonymous OP 3w

Babe…what do you think training really means? How do you think ai generates voices? It doesn’t come out of thin air, and it doesn’t come from paid voice actors

2

Anonymous #1 3w

What do YOU think training means? I work with ai professionally, lol. The voices Duolingo uses are the same voices they’ve been using for decades, and it would be weird for them to change that now that people are familiar with them. They’re not generic text-to-speech, they’re models built from professional voice performers’ voices. Again, the training they’re talking about is for conversational models. It’s a large language model, not a voice model. They already had voice models.

-1