Watkins, speaking to PCGamesN, elaborated that the text-to-speech always starts with a voice actor: "It's part of their contract that we use it [AI] for this purpose, and that allows us to do things like our ping system, where it's capable of saying every single item name, every single location name, and compass directions. That's how we can get that without needing to have someone come in every time we create a new item for the game."
So no, they are not "stealing" voices. Their contracts explicity states that they are training the model. So they are getting paid, which in conclusion rules out "stealing".
Also, from your video:
But to my understanding
Rarely ever good if a sentence starts with "To my understanding"
the AI tools Embark uses to then synthesize the rest of the performance come from models that are trained on millions of other voice actors that have been stolen from in the way that all generative AI models steal from artists.
No, that's the whole point of models that are trained on a single voice - you do NOT use other voice actors because that would completely muddy the voice. The models are trained on a singular voice to mimic that person perfectly. Using other voices is like asking someone to cook a potato soup for you and then you toss in tomato and paprika.
AI is a tool, and a good one if it's used properly. And this is definitely a good use case.
If you really want to make a point against AI, stop using windows as they are trying to push AI bullshit into everything.