Neural networks today are evolving so quickly that their progress sometimes just can’t keep track. They paint pictures, compose music, voice text the voice of any person to list their achievements can be quite long. Another amazing technology we have demonstrated researchers at the University of Washington. They managed to teach a neural network to animate the facial expressions of the former President Barack Obama. And was very believable.
The system needs the audio of the speeches of President Obama and some free time. Scientists pre-fed the neural network several hours of video recordings of the former President, so she could study his facial expressions and identify main areas for further work. And voila! The output we get President Obama, who on camera says the phrase, uttered on some talk show or something. In this case, of course, it is best to see it with my own eyes. So how to explain this in words – not an easy task.
Assistant Professor Washington University Ira Kemelmacher-Shlizerman explains that this technology can be useful in several areas. For example, in virtual reality, human speech can be almost perfectly ported to his computer avatar. This approach can seriously make life easier for the developers of video games that use to capture gestures very complex and expensive technology. But here you just need to record the voice of a man and feed it to the entry of a neural network.
The system is still far from perfect. Because in advance it is necessary to study carefully the facial expressions of the person, and this would take several hours of video recording. But the creators of the technology believe that in the future the process will be greatly simplified. This technology in more detail will demonstrate to the public the exhibition SIGGRAPH 2017, which will start next month.
The neural network learned to animate Barack Obama
Sergey Grey