Hacker News Reader

> We recognize that GPT-4o’s audio modalities present a variety of novel risks

> For example, at launch, audio outputs will be limited to a selection of preset voices and will abide by our existing safety policies.

I wonder if they’ll ever allow truly custom voices from audio samples.

I think the issue there is less of a technical one and more of an issue with deepfakes and copyright

It might be possible to prove that I control my voice, or that of a given audio sample. For example by saying specific words on demand.

But yeah I see how they’d be blamed if anything went wrong, which it almost certainly would in some cases.