MoreLuckThanSkill wrote on May 30, 2023, 10:14:
This sounds like shit, and the actual (written) dialogue seems so generic as to be insulting (was that AI generated also?) but I'm sure it'll be in every game the next few years.

The most important thing about this demo is that they filmed it in 32:9, support those big monitors, companies!

Because whatever voice synthesis they're using isn't great, Elevenlabs AI voices are much better with this with more range of emotion. They are most likely not paying for that here.

Also, none of this is written dialogue. The "AI" generates the lines based off of knowledge it has, and the lines are actually SAID by the player. So you can ask the AI whatever you want instead of being forced to pick from preset choices, and it'll answer the best it can. Also, this doesn't preclude writers from inserting their own lines into key points either.

This is honestly VERY promising tech, and I don't know how any RPG player couldn't be excited about it.

MoreLuckThanSkill wrote on May 30, 2023, 10:14:
This sounds like shit, and the actual (written) dialogue seems so generic as to be insulting (was that AI generated also?) but I'm sure it'll be in every game the next few years.

Just to be clear, the whole idea is for there to be no written dialogue involved. Those aren't pre-recorded voice lines for a main character, that's supposed to be the actual player speaking into a microphone. So in real time, your own words are transcribed to text (obviously, some games could allow you to simply type what you want to say), that text is fed into the AI conversation model to generate a text response, audible speech is created from that text response, and finally that audio is used to dynamically animate the NPC's mouth while it's played back.

The only things the developers specify are certain details related to the character's background and the state of the game world around them, to hone their personality and what they should be more likely to talk about. I.e. this guy slinging ramen tells you about the crime problem and mentions the name of the local gang boss, which would trigger a quest for you to seek him out. Sounds like there's going to be a lot of leeway in how it's implemented, which will have a big impact on how well it works (in the context of NPCs performing their expected functions) and how well it's received by players.

Jackplug wrote on May 30, 2023, 14:42:
To be all honest there was no real difference between that and scripted speech.

It's a bit worse, but also, we hit a point somewhere 5 or 8 years ago where we stopped being able to choose what we say to NPCs. For a while, we had at least 4 options. Then, it was either "angry/sad/supportive/happy," or worse, giving us 4 options that we chose from a single word that often wasn't the tone we thought it was.

For some games, being able to interrogate NPCs in our own words will make up for an deficiencies in delivery. Not every game, not most games, but for the games it does, it'll be great.

Beamer wrote on May 30, 2023, 10:01:

Xero wrote on May 30, 2023, 09:15:

Dev wrote on May 29, 2023, 22:55:
Jeez, I wish I had invested in nvidia 8 months ago, could have quadrupled my money.

My buddy just did. A few days ago his $400 he put in grew to $5000 overnight. Crazy.

I have something similar, but it's still not making up for my Roku stock. Bought that shiz at $32 and watched it go up to $450. Kept buying some along the way. Ended up seeing it grow 8x, only to now actually be underwater on it.
NVIDIA I bought at $34, but bought way less of it than Roku.

All of this is why I don't buy individual stock anymore, outside of a small account I haven't much touched since 2020. It's my "fun" account. Investments are just ETFs.

Stock's path took a twist,
Roku's gamble, friend amiss,
Six feet deep in risk.

To be all honest there was no real difference between that and scripted speech.

This sounds like shit, and the actual (written) dialogue seems so generic as to be insulting (was that AI generated also?) but I'm sure it'll be in every game the next few years.

The most important thing about this demo is that they filmed it in 32:9, support those big monitors, companies!

Xero wrote on May 30, 2023, 09:15:

Dev wrote on May 29, 2023, 22:55:
Jeez, I wish I had invested in nvidia 8 months ago, could have quadrupled my money.

My buddy just did. A few days ago his $400 he put in grew to $5000 overnight. Crazy.

I have something similar, but it's still not making up for my Roku stock. Bought that shiz at $32 and watched it go up to $450. Kept buying some along the way. Ended up seeing it grow 8x, only to now actually be underwater on it.
NVIDIA I bought at $34, but bought way less of it than Roku.

All of this is why I don't buy individual stock anymore, outside of a small account I haven't much touched since 2020. It's my "fun" account. Investments are just ETFs.

Dev wrote on May 29, 2023, 22:55:
Jeez, I wish I had invested in nvidia 8 months ago, could have quadrupled my money.

My buddy just did. A few days ago his $400 he put in grew to $5000 overnight. Crazy.

Yeah, "convey emotion" is going to be the next step. Right now a lot of games use face mapping to do this and the results look really good, even when spoken in another language. The lips don't match up but the emotion in the voice acting does.

They need another AI model to model what the emotional state is of a character at any point in the dialogue and then metahuman and whatever they used for the voice generation needs to be enhanced to make use of that data.

The animation on the face is so bad and the fact he doesnt move around with his task. They require the ability to give it a default "job" animation and then transition into looking up and talking to the player or something man I dont know why they showed this it does more harm than good - its not ready for use.

Jeez, I wish I had invested in nvidia 8 months ago, could have quadrupled my money.

Why do I want artificailly dumb conversations when I can have naturally dumb conversations?

What nobody has learned yet during dialogue, is that the mouth doesn't move so emphatically and dramatic as they always show in games. It's over acted and it makes it look more fake. The lips should move more subtle and the mouth doesn't have to over open when pronouncing words. It's like, they're trying to force make you believe that it's legit. Just look at when he says the words, "city" and "side". His mouth nearly drops.

jacobvandy wrote on May 29, 2023, 18:07:
C'mon, you crotchety bastards, remember when text-to-speech used to sound like the computer from WarGames and pronounced half the words you tried to type in as a sequence of letters? The actual tone of the voice can and will be improved as well as customized in many different ways, I'm sure, but how this works under the hood and what it means for indie devs who don't have the budget for writing a million lines of dialogue and voice actors to deliver them seems pretty amazing.

Beyond that, it gives flexibility. Characters can still be written, but also flexible.

Take Oblivion. Ancient now, but I adored how janky it was with it's single voice actor for each race/gender. Imagine a character gives you a quest. Has 3 lines written. You do those 3 lines. But you're still curious, so you probe. Maybe you want to know what they think about other characters. Or if they've heard about the weird bunch of robberies no one can solve. Or why they're talking to you like you're a nobody even though you're the head of the magicians and warriors guilds and carrying a sword worth more than their house.

Or, you just want to go through the quest and hear the basic lines and go on your way. Really depends upon the game.

But if you're someone that wants "immersion," having characters adapt to your past deads would be cool. Did you just murder a whole town? Maybe they should treat you differently than had you saved it.

jacobvandy wrote on May 29, 2023, 18:07:
C'mon, you crotchety bastards, remember when text-to-speech used to sound like the computer from WarGames and pronounced half the words you tried to type in as a sequence of letters?

Of course, and most text to speech still does even if the audio quality has improved. Diction and pronunciation is still well beyond the grasp of what is available currently and will be for some time. Modeling natural language is hard. Try speaking without inflection, contractions, accent, or even the way your mouth forms words (which is unique to every human). You can't. That's why this comes across so poorly. It has none of the audio artifacts we're all so used to hearing and subconsciously processing.

jacobvandy wrote on May 29, 2023, 18:07:
The actual tone of the voice can and will be improved as well as customized in many different ways, I'm sure, but how this works under the hood and what it means for indie devs who don't have the budget for writing a million lines of dialogue and voice actors to deliver them seems pretty amazing.

The tone is irrelevant. The problem is that it is flat. It's something you sort of recognize as speech but your brain already tags it as artificial and so rejects it as unnatural on a subconscious level. You can say "they're totally going to fix that" but that's not guaranteed. That's wishful thinking, not a statement of fact.

There's nothing "amazing" about overhyped buzzwords that are trying to push premature solutions in search of a problem. It's like Bitcoin in that regard. All the same hype, the same buzzwords, and the same promises. Yet 14 years later, Bitcoin is a worthless joke to all but the most delusional. I'm not paying my utility bill, buying groceries, or receiving my salary in it. Are you? Probably not.

The backlash against "AI" is already happening because of the same overhype and the same lack of practical use to the average individual. If, and that is a mighty big "if", and when that changes, it won't be some sweeping, revolutionary change. It'll be a long progression of incremental improvements in already existing products and services.

Until then, this is just yet another example of a chatbot that can read a string field.

this video can be any pre-recorded video demo....

I could pass on the speech option. Just free type questions and the AI comes up with a response trained from the game lore and story. If it gets long winded, there should a be a button to interrupt and then you can just say, cut to the chase and cut the bullshit and depending on the NPC's disposition it can either answer, take offense and tell you to rightfully bugger off.

C'mon, you crotchety bastards, remember when text-to-speech used to sound like the computer from WarGames and pronounced half the words you tried to type in as a sequence of letters? The actual tone of the voice can and will be improved as well as customized in many different ways, I'm sure, but how this works under the hood and what it means for indie devs who don't have the budget for writing a million lines of dialogue and voice actors to deliver them seems pretty amazing.

Except for a few special cases, like if the AI is generating dialogue for a robot character, I think it will still be decades before we get to a point where this kind of thing is indistinguishable from half-decent voice acting with a decent written script. There's so much more to it than just reading the words. It's a performance. Good luck getting all of that nuance out of AI. Even if this is just used for background NPC chatter, it would stick out like a sore thumb. However, this would be great for games like Detroit Become Human and otehr games where robots play a large role.

Rough, but I see a lot of potential here.

All the stuff in this video that could be painstakingly hand-crafted has been moving towards automation for many years. If these tools are used well, the results could be great. It'll still take a lot of hard work by humans to make that happen.

Sadly, when it comes to character design, dialogue, and general writing, human writers have been performing extremely poorly for a long time. It shouldn't be that way, but as a result I wouldn't be surprised to see generative AI do a better job in the short term.

Hopefully someone will at least bother to make it use contractions properly someday. "I am worried about the crime around here" when spoken just screams "soulless robot".