Modders are using AI to put voice acting in Morrowind
pcgamer.com
pcgamer.com
But this had the effect of dramatically reducing the number of options when talking to NPCs and also creating some immersion breaking artifacts like an NPC sadly recounting a personal tragedy and then immediately cheerfully discussing the weather or whatever.
It's also strange when the same four voice actors play so many characters. It's just tedious and immersion breaking.
AI has the potential to improve this situation in several ways.
It's a lovely redeeming quality of both games, by far. Having the same voice actor for everyone is absolutely freaking hilarious, too, to boot.
I mean, how can you not love this lol: https://youtu.be/N6hVmn9FM7o
In something of the caliber of TES they could just get a few hundred voice actors. It's something like 100m budget vs 1b in revenues.
It completely changes your workflow because you need to have your script ready and finalized to record all the lines all at once. Say good bye to iterative development. Want to tweak something or add additional lines? Yeah, that is going to cost you.
I think "AI" based voice generation (well data driven voice generation, I will die on the hill that it should not be called AI as it is not intelligent) will absolutely have a place for prototyping games and having filler voice acting.
The problem is though, that good VA can absolutely make or break a game. Will players really tolerate "AI" voices? I hope not! There is a reason certain famous voice actors are paid good money: it is an art.
People will watch certain shows or play certain video games simply because they have a voice actor that they like in. You can't get that emotional attachment form generated voices.
Voice carries so much more information than just the text. It gives hints to the emotional state of the speaker, their background and education, their intentions. "AI" can never replace top-tier voice acting from humans simply because that would require a general intelligence.
Vocaloid managed to turn people’s emotional attachment to generated voices into a billion-dollar industry
(I’m actually crying right now, just remembering the emotional impact of The Disappearance of Hatsune Miku — a song written from the perspective of a piece of voice synthesis software which is in the process of being uninstalled, and performed at a speed no human singer could match, as she tries to squeeze in one last goodbye T_T)
This exact same quote has been used as a desperate means of gatekeeping for what it means to be intelligent across every domain. Deep blue with chess, AlphaGo, diffusion generative art, the list goes on and on. It's amazing that people continue to make this mistake to me.
Everyone thinks that their domain is safe, and everyone is inevitably wrong.
There have already been many high profile realistic sounding TTS engines in the last 10 years. Some of the newest advancements allow you to combine a TTS voice with a recording of your own voice. So a game developer can record a line of dialogue angrily and with emotion, and then this is transferred onto the appropriate TTS generated voice (elf princess, dwarven blacksmith, etc).
Voice acting is a field absolutely ripe for disruption.
https://mashable.com/article/james-earl-jones-gives-rights-t...
There are certainly people who do this, but you're projecting. This is objectively a minority of the target market. I could canvas 100 children who have watched Amphibia or owlhouse and it's doubtful a single one of them could name a single voice actor in those shows.
This applies even more so to voice acting in video games, I'm an avid game player everything from GTA, to tomb raider, to portal, to baldurs gate. Gun to my head and I couldn't name a single person who voiced a single line.
I know it's just nostalgia talking, but I feel like the text dialogue was one of the things that made MW as special as it was. At a certain point I think text becomes easier for a user to digest - you don't necessarily want some NPC talking at you for ages about the history of Vvardenfell. And MW was very rich in history like that.
No to mention, this can sometimes become immersion-breaking. Especially with such data dumps, voice actors would forget (or not be informed) about the context in which their lines are being heard by the player, and then you end up with a random villager or space soldier suddenly giving you a TED talk.
I love immersive sim games. Miss back in the day when we didn't have maps for dungeons and you were supposed to be drawing a map on paper. I never did, as a little kid I just learned how to memorize dungeon layouts really well and now I memorize any place I go on foot.
It has a nice ring to it!
The gameplay, especially from a melee perspective, is rough at best, and directly unfun at its worst.
No sprinting, no mounts, no fast-travel-from-anywhere (beyond mark/recall and the teleport to temple spells) and the in-game fast travel systems were varied, thought out and well placed. Having to plan a route of mage guild teleports, boat rides, stilt striders, and/or the teleportation complexes helped immerse the player in the world.
Boots of blinding speed to the rescue! No idea why authors think it was good idea to make default speed so low and lock game-fixing item behind some random-looking quest in the middle of nowhere
The combat in this game was considered rough when it was new, and it has only aged like milk since then.
Imagine you made that statement on May 27, 2019, the day before Outer Wilds came out.
If all you look at is AAA experiences, then maybe I can understand this viewpoint, but this is absolutely not true in general.
Given that movie theaters sprang up in the early 20th century, [2] you could make the argument that video games haven't even reached their equivalent of the French New Wave yet. Given how far someone can go with a little bit of self-taught asset creation, I think the best video games will be made after you and I have passed away.
So hard disagree there.
Arena and Daggerfall (the two predecessor games) were procedurally generated, and just felt "more of the same" after a while. My understanding is that the world in Morrowind was lovingly crafted by hand. This might have resulted in a small in-game world—but each inch of that world felt special.
Interestingly, the Witcher 3 is probably the one other game that gave me the same "this world is handcrafted" feel (dunno how that was built though).
The illusion starts to fade when you wander further away from the school, but the game smartly keeps most of its critical path around Hogwarts and Hogsmeade.
Ever play Riven? I remember reading that "every pixel was crafted by hand" (which probably wasn't quite true, but, probably not enough to matter. That game was a work of art)
(Dungeon Keeper -> Dungeons 2; more than a straight remake)
(Master of Magic; dec last year; nearly an exact remake - switch to hex grid, few other tweaks, maybe more down the road :hopes:)
(Pharaoh, A New Era - tomorrow, possibly an exact remake?)
I cannot damn wait until the Banjo-Kazooie decompilation project is complete and someone starts working on a Render96-esque version of it.
On one hand, you've got flawless voice scam calls, but on the other? I can imagine a new style of fan-edit movies, but without any limitations of having to splice together film and audio from prequels/sequels/etc.
We're in for a wild ride.
As someone who made this as a sideproject, I can totally see how if you have a team of people behind a project for a game like Morrowind you could go deeper and create whole narratives and quest systems around these "ideas" - without the need for 10s of voice actors and game-modders
Quite fun for those who like, have been playing the same game for 10+ years!
I've even used Japanese inputs and the outputs kind of sound like the originals speaking native English. Still a bit uncanny and less stable than inputs that match the model's language.
Forget deepfaking, it's quite literally a bottomless pit of unique voices that don't exist.
several other ramifications of this is that file sizes can actually be smaller again. Most/significant part of the file size in games is audio at various formats and high qualities. But since pre-recorded can go away, it can be replaced with prompts tied to some NPCs that augment their response. The training sets can just be the file, and the training sets (edit: the MODEL) can actually come with the OS or be a shared resource on the system, instead of packaged with the game. Smaller additional training sets (edit: the model, not the training set) can come with the game.
I have a different idea regarding compensation paths for human beings, but its not likely that it will be followed, or lucrative even if implemented.
You might tangentially download the code to train a new model from scratch or update the model you received.
The important bit is that the training data is HUGE and the model is not.
I respectfully disagree. One only has to look at Mass Effect's voice acting for male and female Shepard. It's (almost) exactly the same dialog lines, both performed by humans, but Jennifer Hale just kills it. The voice alone made a lot of players think of Femshep as the canon (despite the male model being present in the box art and most PR media).
I do agree in that we might need less voice actors - but mostly for non-key NPCs. Humans will still play a role.
On that same note, it would not surprise me if non-key NPC's dialog was not written by a human any more, but by a machine. A human "writer" could be replaced by a "chatgpt prompt engineer". But again that will only apply to secondary NPCs.
https://www.instagram.com/reel/CoVPCo2sMIj/?igshid=YmMyMTA2M...
You can already feed this generated responses to say back.
In the near future, humans will play a role, for 5 minutes. We agree on that point. In the also near future, we'll have enough samples for a diverse audience.
Its gonna be a some time before the "baseline" console/PC is fast enough to run a game, language model and voice model all at the same time.
And it may take even longer for software support to be put in place and for devs to actually use it. Unity doesn't even natively support Opus right now, which would dramatically reduce VA filesizes
for example:
- simply pre-rendering scripted responses using voice models. same audio files, large games. no voice actors.
- the same thing without QA even checking if the voices sound good
- larger scripts but internet connected doing the voices in the cloud
- language and voice models, internet connected rendered at the data center
- a smaller set of prerendered voices offline, a larger set when internet connected
- all coming client side
No more generic fetch quests. No more dungeon grinding. Adventures only, please!