I don't usually write articles around my mods. In fact, I *never* write articles around my mods. Does anyone?

I work in tech for my day job, and, having spent alot of time with ElevenAI specifically, am in an interesting position to share my thoughts around AI - specifically voice/sound AI, and the criticisms and concerns I see coming up in the comments section and in the discourse around this mod and others like it.

I've written the below not to attack those people, or defend the ethics of voice AI mods like this one - but just to put my own thoughts down in writing, as people clearly have an interest in the subject.


Concern 1: But didya get permission to steal the voice from Jeff Baker's mouth?

Mods have always existed in a muddy legal territory. 

The game ships with a load of assets - meshes, textures, sounds, music.

People have come to accept that the remixing of those original assets - even if you don't make anything 'original' yourself (we'll get to, later, if this mod does have anything original in it) - like a modder who kitbashes a new armour together from existing meshes and textures and releases it as 'Redoran dust warrior armour' or something. Or you do an acoustic guitar cover of the Morrowind theme song for a bard to play in a tavern. Or you re-use the bug texture from a robe and use it for a creature. 

No one worries about getting permission from the original modelling team to make that armour. No one worries that they're putting them out of a job.

Morrowind is a game that's over 20 years old now - no one is going to be missing out on royalty checks from this content. 

If anything, if this mod was to become incredibly popular (be still my beating heart!) it could actually push people to purchase Morrowind, either to install again or for the first time - actually making Bethesda and crew money, and sustaining fan interest in the series until the next Elder Scrolls game.


Concern 2: Even if it's not unethical, the results are creepy and unnerving. 

This is a really interesting one. In part, I think it's because it's new technology, and so we're perhaps a little *too* impressed by the results. Voices are just made of composite elements - pitch, volume, reverb, etc - with some patterns for how we annunciate or pause in our speech. It's perhaps no more complex than the disparate elements of say, an armour mod - it's just less well understood by players as a whole. When we first saw normal maps and diffuse maps and all these cool texture variants that simulate slime or metal and light hitting surfaces, we thought it was witchcraft. Now we just see being able to simulate cloth physics and hair whipping in the wind as par for the course for our games. 

Perhaps the other problem is the fact the voice is such an intimate thing. A texture or a model is one thing, but a voice - mimicing it, replicating it - feels akin to copying someone's soul. When you create an armour that looks realistic, people clap. But when you create a voice that fools people into thinking its real, they get creeped out. I definitely did when first making this mod. But it quickly passed, and was replaced with a sense of wonder, and a feeling of 'wow, I can really make Morrowind feel like a more immersive, reactive place for players to have fun in with this tech.' 

Once you get over the fact that it's possible, and passes for the real thing, I think the feeling passes. 


Concern 3: It's too easy to make this stuff. You just press a button and the AI makes a voice. 

This is both true and not true. It is true that you can get 70% decent results from ElevenAI with little effort. But in order to get those 95% immersive results I'm trying to do, it's much harder. This includes:

Endless 'regenerating' of the lines for it to put emphasis in the right place. The number of times I've had a take that has 'You'd better watch yourself' delivered in a threatening baritone, just to have the second part 'or I'll call the guards' delivered in a cabaret sing-song that ruins the whole thing, I can't count.

Re-writing of the original lines. Morrowind wasn't written in a way people actually speak. The greetings were written to have a wikipedia style hyperlinking of topics and phrases. Ie - I'm Von Django, the modder for Bethesda's award winning game Morrowind. Many of these lines have been re-written, shortened, and humanised to work as delivered speech.

Re-writing phoentic spelling of Dwemer ruins, local alcohols, or Dunmer language. 'Lahloo, Must-ay,' etc

Originally, the scope of this mod was just to make quest-givers awknowledge the quest you were on. But this mod has expanded to include so many new things, like custom idle thoughts for some NPCs, voices for guild guides who you can travel with, rumours for cities and towns, and even some wild stuff like Galbadier actually having a conversation downstairs in the Mages Guild while she's distracted and you're on a mission for Ajira.

This is where the 'original' part of the mod comes from, beyond just pressing the button in ElevenAI and getting a sound file.


Concern 4: The mod isn't the problem. The problem is the implication of the tech for voice actors, and humanity as whole. 

I'm definitely not qualified to talk about the impact of AI on humanity as a whole.

It's certainly true that some technologies often put people out of business. Photography, although it didn't kill convential art, probably put alot of portrait painters out of business. Or perhaps allowed people who could never afford to have their portrait a taste of a similar thing.

Having used ElevenAI extensively, I don't think AI voice models will put voice actors out of business.

There will probably be a shutterstock bank of AI Voices for games that can't, or won't afford, to pay for original voice acting. 

But this will never be able to replace the impact of the real thing. 

As someone who has spent too much time with it - the only reason ElevenAI works here is a very limited application - short, sharp single lines of dialogue. ElevenAI is incapable of delivering a stirring speech, much less a realistic exchange of dialogue between two actors. ElevenAI doesn't understand the real emotional impact of what it's mimicing, and certainly can't capture the nuance of two actors bouncing off each other. It can only accidentally hit the right notes, in the right order, for a very short period of time - if you make it reroll enough. It's the equivilent of 5000 monkeys on typewriters eventually writing Shakespeare.

Players will know when they are in an AI voice acted game (when you have full, deep back and forth conversations) - and will always prefer the impact of the real thing.

I'm sure there will be AI addendums to actors contracts so the studio can use AI to record additional lines for new content as the game is developed, or additional responses to dialogue trees if they come up with additional ideas after the recording session. But it'll be, in the main, a backup plan to give studios extra leeway in development rather than a replacement for the real thing. 

Of course I could be wrong. But as someone who has used this technology a huge amount to make this mod, I think people have less reason to be concerned about this technology than this mod may make them feel.

It's a cool trick. But it's not a game-changer.