Have you been looking at AI-generated voices for third-tier NPCs like the ATC? Google's latest AI voice tech sounds amazing.asked by Shaney

Yes, that is definitely a very good point. We can do that for NPCs. Mostly our target has been really for the system voices. I know there's a question down here on ship computer voices, so I'll save that for later, but a lot of system voices, especially with a lot of incremental game design: when we want to go record and cast an actor and get them in, we have an amount of lines for them, everything makes sense, we go in there and get enough content. But at some point later, you'll want to have this new gameplay feature, and that thing only has five lines, and you're like, okay, we now need to call this actor up to have them come back in and record five lines. Or if it's a work-from-home studio, whatever it is, they have basic minimum fees for sessions, and it's just not financially viable. So, especially for our iterative game development, we want to look at these AI-created voices and whatnot. But we also would like to be able to still pick and choose our talent, pick and choose the people we want to have in our game. And so we're looking at ways to be able to create our own voice packs and say, actor so-and-so, you're going to be the voice of this ship and we're going to go capture your essence, and be able to create a voice from you that we can then drive with our text-to-speech. Or speech-to-speech is another thing that we've been looking into too, where we can take a voice actor, and staff or anybody else can read off those lines in the way we like, but it'll sound like that person that we captured. And it's another way to bolster content without having to hire talent for just five lines a piece. But obviously we still pay them proper buyouts and proper licenses for their voice. We do make sure that the talent are treated well and that we do all the right legal things with that. It has been adopted in the industry. We're not the only ones to consider that. Film is also looking at ways to do better ADR. You have a project, a film, and there are requirements to say, oh, you can only ADR 10% with these speech-to-speech things, and then you're able to bolster it without having the guy come in for five lines or something. And that's a benefit. But we obviously are very keen on that, and we want to push forward with these AI or text-to-speech created voices and really have a lot more content to work with.

There are technical issues that need to be overcome, though, because I know some of these AI-generated speech samples do sound amazing, but a lot of that AI is trained on natural language and is trained on known language. And a lot of our language in lore isn't known and is a little bit tricky for some of these things. And it's really important that we make sure we don't end up in the uncanny valley, because I think when these AI generations get something wrong, you can really hear that it's not a person speaking. So yeah, it's a lot of research.