AI Voice Demo Game (Open in Quest Browser)
WebXR Microphone Troubleshooting Guide
AI voice support is complete. Server is located in Dallas. Tell me about your experiences in the comment.
To use AI voice feature, activate speech function on the start screen.
Voice Chat Demo in Quest
You can talk without the speech input box. (Note 3.2.1 next to the send button.) But I recommend you leave the Specch Menu on. Streaming or speech recognition fails often. You’d want to know what’s going on.
Action Speech Demo
My thoughts(or skepticism) about AI
- The worst part is latency. It takes up to 5 seconds to generate a conversation. It breaks the immersion. -> This issue is partly resolved by streaming. But sometimes a huge lag is still happening with streaming.
- The second-worst part is the TTS. It never feels human, and it sounds stupid too often.
- AI itself doesn’t make a content in 3d games.
AI language model is more suited for text chat services.
This can still be effective as an advertising scheme. I’m mainly doing this for advertising purposes.
For a longer user experience, it’s necessary to save conversation and let the AI remember the dialog. And maybe let the users customize the personality profiles or other prompts.
That’s what AI companion apps like Replika are doing.
It will depend on user engagement. If this gets some use, I may apply it to all of the VR Hand Revolution games. I may even add more postures suitable for conversations.
Anyway, my priority is implementing those with 3D-scanned pornstars.
Graphically, it looks a lot better now that it has lipsync. Feature-wise, I think the most important fix was streaming.
Tell me about your experiences with latency. I’m really curious how it works for US and European users. *I’m connecting from East Asia, so I can only see the worst latency.
Seems like it’s finally time to open the 3d scan files. 🙂
Hey, Rockhard, great idea, to give the charackters some voice, just curious, why dont u just cache some of the AI responses, especially the basic one coming on the action, since there are not that much actions, mb having like 20 generated responses per action cached and then randomly played, and let say every 50 action u request a new thing from real AI (and cache this into the buffer for this action) to give it some more live. Also not sure if u using own AI or the requests directly, but they are AI voice covers and i think u could replicate the original character voice. The request mostly costs a bit of something, but with the proper Buffer (u can buffer the audio directly in that case) its gonna be not so expensive i assume.
I’m using OpenAI and Azure Speech, so every call costs. Not many users are using this. Cost is minimal, it’s okay for now.
AI is repeating almost the same words since there’s not much to say. 🙂 It’ll be no difference if I cache all of it.
I think you mentioned training TTS with custom voice sets. I wonder if they will support emotions if I train voices.
I saw several Skyrim mod videos. All of them spoke in a single tone of voice without any emotions. It sounded dull.
Currently, Azure provides five sets of stock voices with ten emotions each. And all of them have pitch and speed controls. It is stable and well documented.
Finally, the AI voice chat is added. This is nothing new. It’s too common these days. You will find better, easier solutions on smartphones.
This voice chat feature could be useful to some people depending on individual tastes.
Personally, I like the reaction talks better. Although the voices sound like machines.
What would it take to add a feature where the cum shot stays on the model’s face in pose simulator? Similar to how it appears in the Faye elevator comic.
I’ll check that after I take care of the current issues. It doesn’t sound too difficult, but I can’t really know until I try it.