Jul 07, 2022
In General Discussions
On November 6, 2020, South Korea's MBN TV station successfully broadcast the main news of the day using an AI anchor based on the real anchor "Jin Zhu Xia". Because the image of the AI anchor reporting the news is very realistic, it immediately caused a heated discussion among the public. In the film, the left is the live anchor, and the right is the AI anchor: News report: Central News Agency "South Korea's first AI anchor goes to work, will it help or pressure the journalism industry (film) " New head shell < too realistic! South Korea's first AI anchor makes an appearance, and netizens call it "creepy " Can the current AI anchor completely replace the real anchor? What are the features and limitations of this technology? We invite experts to clarify. Let's see what the experts say. Xu Zhizhong (Assistant Professor, Department of Information Management, National Pingtung University of Science and remove background from image Technology) November 25, 2020 This AI news anchor is mainly developed by South Korea's MoneyBrain company, and has not disclosed technical details. However, from an academic point of view, it is not technically difficult to synthesize fake AI anchors. For example, film generation and sound synthesis, there are open source codes and research papers in related research fields. available for reference. The easiest way to make an AI anchor is to input the relevant press release into the computer, shoot a live-action video, and apply the face-changing technology. Therefore, it is not difficult to produce similar results if the academic and research teams in Taiwan have expertise in combining video and audio processing. In fact, in the field of image and video processing, the most common traditional face-changing technologies, such as the well-known DeepFake deepfake, or other open-source CycleGAN (Note 1) technologies, have the ability to synthesize the above results. Although there is little mention of the ability to generate AI anchors with realistic visual effects and voice directly from the manuscript in the current related literature, all relevant technologies in each part are of a certain level.