3 ms·
nice, glad to see someone is already on it! I had opined that video games won't need voice actors any more and way more fleshed out dialogue. A random NPC runni
by yieldcrv 4y ago
nice, glad to see someone is already on it! I had opined that video games won't need voice actors any more and way more fleshed out dialogue. A random NPC running around in Fallout in the distance? You can have a full conversation about some relevant contexts in the world, with spoken responses.
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.
- oneoff786 4y agoYou don’t need the training set available to run the model.
- yieldcrv 4y agoyou're right, I should say just the model, or trained model. what terminology would you use? For example, with stable diffusion I download a.... model? and can also train a model?
- oneoff786 4y agoIt’s just the model. 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.
- otikik 4y ago> video games won't need voice actors any more and way more fleshed out dialogue 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.
- coldacid 4y agoHonestly, I think we need more voice actors, not less. One of my biggest complaints with the dialogue in Oblivion wasn't the inanity of the "conversations" but the fact that there weren't enough different voices to go around so everyone of same race and sex would sound the same.
- yieldcrv 4y agoI believe that the only reason you would write this is because you haven't seen the services that let you train a voice generating AI from a few minutes of your own voice or from a few minutes of another person's audio. https://www.instagram.com/reel/CoVPCo2sMIj/?igshid=YmMyMTA2M2Y= 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.
- otikik 4y agoMy point was that a human will always be able to convey more context via tone and inflection than a machine. Machines will need to use manual context tags like <angry>. At which point you might as well just hire a person to talk to a mike. Machines also noticeably struggle with more nuanced emotions like sarcasm, or disdain, even with tags. That’s all well and good for something like merchants. It’s not good for main characters.
- 127 4y agoI don't really agree. Femshep is way overhyped and maleshep is just fine, if not even more iconic. Femshep sounds like trying too hard to come across as gruff and experienced military veteran and it just doesn't work for me. Comes across as fake.
- punchclockhero 4y agoStarted playing Mass Effect with Knights of the Old Republic fresh in my mind back in the day. It was so jarring hearing the same voices with different characters.
- brucethemoose2 4y agoPrerecording VA with AI is one thing (and is super cool and practical right now), generating it on the fly is a whole different can of worms. 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
- yieldcrv 4y agookay there are many baby steps before this future 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
- BizarroLand 4y agoWhat I picture would be a Diablo style open world MMORPG with NPCs voiced by AI, quests generated by an AI that has information about the world as it has progressed over time, and the quests tailored to each person's character and play individually. No more generic fetch quests. No more dungeon grinding. Adventures only, please!