Digital avatars and the potential for virtual immortality
A research team from the School of Computing at the University of Leeds has received significant media interest following the development of digital avatars which show the potential for one day rendering a person virtually immortal.
This collaborative research between Dr James Charles, Dr Derek Magee and Professor of Artificial Intelligence David Hogg was based on a continuation of work originally conducted in the late nineties by Professor Hogg’s research group. It also includes work carried out while Professor Hogg was a Visiting Professor at the Massachusetts Institute of Technology (MIT).
Professor Hogg’s research is in the field of computer vision and its application to human-computer interaction and scene monitoring. These areas of research combined with an interest in devising techniques for learning and then recognising the shapes and behaviours of objects within a scene mean that the research team are perfectly positioned to build a series of algorithms that can recognise and track individual characters and also capture their body language, facial expressions and voice.
Reanimating characters from TV shows
The research team’s goal was to create a system capable of learning to generate new audio-visual content of a chosen TV character from a TV show, in the form of a moving, gesturing and speaking 2D avatar.
By using characters from popular TV shows, the research team has been able to use the unconstrained pre-recorded data from boxsets to create computational models. A machine learning tool analysed 97 hours of television show scripts to understand how each character typically puts words together. This enabled the research team to build virtually talking interactive avatars of those characters who are then able to generate new and interactive content of their own, seemingly rendering the person virtually immortal.
Although the software is only a prototype and the rendering of the avatars needs improvement, the researchers intend that the work will create a natural interface between humans and technology with the potential to put a face and personality to voice-only assistants. The avatars could also interact with each other. The research might also be used as a way of generating ground truth audio-visual data for training Artificial Intelligence interactive systems
Professor Hogg will discuss and explore ideas associated with this fascinating research and related work in the Image Analysis module offered as part of our Advanced Computer Science Masters level courses.
‘The purpose of our research has been to demonstrate the principle of automatically learning interactive behaviour, appearance and speech patterns from the characters in TV boxsets. One day, this could open up a wide range of applications for more realistic and engaging digital avatars, and the intriguing possibility of what we have called “virtual immortality”, a natural next step from paintings, images and video in capturing the true likeness of a person.’ Professor David Hogg