Idk, I guess this could go either way in court but there is a pretty strong case that the purpose of a TTS or voice conversion model is not to imitate someone’s voice. The way this is worded, to me it seems to target services that advertise creating fakes as a feature.

I guess there is a lesser problem since a model that let’s you replicate your own voice or someone’s voice with consent, which can have genuine, moral uses and a model that can replicate a strangers voice is the same. But honestly, seems unlikely that they go after the open models, rather than directly useable applications. And applications can be designed in ways that don’t actively enable abuse, by prompting the users for lines to read rather than letting the user input arbitrary voice snippets.

I agree with the hacker news guy, this comment makes much more sense:

https://old.reddit.com/r/LocalLLaMA/comments/1q7qcux/_/nyhtae5/

If anything, oop feels like a lazy attempt by ai companies to kill any regulation