Virtual 3D AI

[DragonK]

Hi Guys,

This is not a feature request, but more like a wishfull thinking. But just had to share a thought.... If we had something similar for LMCE.....

Part 1 : http://www.youtube.com/watch?v=7W52TL9Akv4  
Part 2 : http://www.youtube.com/watch?v=ZmlrkQBjVok
Part 3 : http://www.youtube.com/watch?v=FxK-pOFFx1o&NR=1&feature=endscreen


Karel

PS. Her name would be Sarah and not Denise....  

[Marie.O]

We are expecting a proof of concept by you by April 23rd, 2012.

[_if_]

wow that would be killer 
propably there is nothing similar as opensource around, right?

[tschak909]

Do you guys truly not understand, that if you don't write it, it doesn't get done, and therefore does not exist?

This is an open source project. Do it.

-Thom

[DragonK]

Thom,

Where would the best place be to try and start. What would you recommend?
I would need ALOT of guidance.

Karel

[grind]

I also like that very much! And the best would be if you could customize the face.

Btw. yours would be Sarah? Maybe because of Eureka?

Nicolai

[WhateverFits]

The face is customizable and it is $100 to $850 per license. It has an API but the biggest integration part comes with the expensive license.

[Sigg3.net]

Just searched for open source 3D talking head and found Xface. The project died in 2008, but they reached version 1.
http://xface.fbk.eu/index.htm

There are probably heaps of other "3D butler" projects out there if you look for it:P

Problems are not tying it into LMCE, but how it would play with other media playing on the instance..

[JaseP]

Blender and Makehuman to make your 3D mesh, then just pre-render video that delivers a video scene for every use case,... Ought to take no more than 50 to 60 GB of hard drive space to store all the animations... And maybe 9 months working 16 hour days to do all the animation.

Sorry for the sarcasm... But, I'm just not all that impressed by this kind of tech. Now, a neural network AI (minus the Final Fantasy-esque digital sex doll) to interface with LinuxMCE,... complete with visual and speech recognition, with predictive capabilities, and maybe a personality,... That would impress me. But we'll have to wait another 15-20 years for that kind of thing.

[Armor Gnome]

I have actually considered doing a similiar idea in my own set up.  My ideas though are much more simple and use looped or very long video files with previously recorded text-to-speech statements.  It would be ugly but just nerdy enough for me to love. The other idea besides a human persona was a robotic face or HAL glowing orb on-screen as the audio is played.

My roadblock to this was time and experience to develop it, the slight load time for xine after receiving a play command and most critical, I couldn't play a "blank" video file and an audio file at the same time with the amount of time I put into this.


Implementation idea if anyone is interested in putting time into this for themselves:

Code: [Select]

Event = press a category button on an orbiter
Response = loop interface video on local on-screen orbiter, play audio file "Here is a list of your videos.  Which would you like me to retrieve for you?", stop looped video

Event = selecting a video file from data grid and then pressing play
Response = loop interface video on local on-screen orbiter, play audio file "Now loading [filename] please wait while I prepare it for you", stop looped video,  play selected video file.
Note that this would not function with UI2 as the selections are not separate screens, I am sure with enough effort it could be incorporated but for the short-term a way to play video and audio at the same time would be a huge step for this.

[_if_]

I've started learning blender some time ago to be able to put together a 3D head for that purpose. It turns out blender being a damn complex program, so I'm still doing babysteps and am not even close to modeling a head but I continue trying.

The plan for the begin was like this:
first learn blender
build a 3D head
use rigging to make the lips move (blender can record my actual lips with a webcam and sync the models lips - once I find out how that works...)
sync to already recorded audio
render animations.
start replacing Sarah in the first step of AVWizard and see if that makes sence
If I ever get to this point, we can think about integrating it into the whole wizard and popup screens like the "you have added a new device..."

I know this is a very hard task and maybe I will give up some time, but at least I'm trying.
What are your thoughts....doable?


IF

[JaseP]

Try MakeHuman;

http://www.makehuman.org/

It creates the mesh, etc. for the whole body, but you can export to Blender to do the animation, just focusing on the head. It also creates controls for animating the face (and body), but those can be challenging to use (like operating a marionette). The hair typically does not export (you probably would need to do the hair yourself, or find a mesh you can import).

[_if_]

cool thanks, I will look into it!

[Armor Gnome]

Sounds good _if_ ,

I hadn't thought about the Sarah wizard since its already pulling from pre-rendered video and audio files.  Let me know how you are coming on this.  It is not something I am still actively working on but it is a cool idea.

[_if_]

Hi,

just wanted to give a little update how far I came until today.

http://youtu.be/uxaLhACVKV4

the whole model may totally change and the lipsync will be improved, as well as the hair (I know you can still see the short hair).
I'm learing blender more and more and will keep improving the whole model when I find time.

Use your imagination and let me know what you guys think.

greetz
IF