Reporting by XinZhiyuan
EDIT: Peaches sleepy
【New Zhiyuan Guide】Daily economic news live broadcast for 70 days, no one found that the anchor is a digital twin virtual person? The world's first digital twins to be indistinguishable from the true and false of real buddhas are N Xiao black and N Xiao Bai. The evolution of the new digital revolution is becoming more and more accessible.
"At present, the company is cooperating with the relevant preliminary work, and the implementation of the commitment is still in progress..."
On December 17, the anchor of the "Daily Economic News", N Xiaohei, was broadcasting the information of the day in a hanging river, and everyone held the mobile phone and watched with great interest, and no one found out - this anchor is "fake".
He is the world's first digital twin N Xiaohei to be indistinguishable from the true and false of the real prototype.
Moreover, not only N Xiao black is virtual, he also has a digital twin partner: N Xiao Bai.
What is even more surprising is that in fact, the two have been working together to broadcast the news for 70 consecutive days, but almost no fans and users have seen it.
Probably because their faces, expressions, body movements, etc., are too real and natural.
Only many people occasionally wonder: "These two anchors are also too dedicated, going to work every day?"
The first batch of digital twin anchors who are difficult to distinguish between true and false have arrived
It wasn't until December 20 that the answer was officially revealed.
After N Xiaohei finished broadcasting financial information, he announced the truth:
"Hello everyone, I'm the AI anchor of the Daily Economic News, I'm not myself, but his digital twin. I've been quietly online and running non-stop for 70 days. Many people think that this is a video recorded in advance by a human anchor, but it is not. Everything you see, including the text I'm talking about, is generated and driven by ai Xiaoice frameworks."
At this time, he still spat out clear and natural words, plus gestures and micro-expressions, which was no different from the state of his real person in the Douyin number "N Little Black Finance".
The "maker" behind them, Xiaoice Company, has also officially announced that a new digital twin based on Xiaoice deep neural network rendering (X Neural Rendering) is coming.
This is the first time they have raised the overall naturalness of artificial intelligence virtual humans to a level that is almost indistinguishable from real humans.
Xiaoice and daily economic news jointly created digital twin anchors N Xiaohei and N Xiaobai, also launched online with "Every AI TV" on the same day.
It is worth mentioning that Xiaoice company has also realized the unmanned operation of the whole process of video acquisition, editing and broadcasting for the first time.
Based on this progress, "Per AI TV" has also become the world's first 7x24-hour UNinterrupted AI video live broadcast product.
However, netizens are also well-informed, and there will naturally be no less doubts.
For example, AI anchors have seen strange things, Xinhua News Agency, CCTV have done, what is special about these two new "virtual anchors"?
What progress has the new technology brought? And how? How will it change our future?
Next, let's go a step further.
Digital twin anchors, what new technologies will they bring?
First of all, let's briefly talk about the birth process of these two "virtual anchors".
First of all, the training data of the two virtual anchors is from the well-known financial anchors N Xiaohei (N Xiaohei Finance) and N Xiaobai (per Xiaobai Fund).
Then, combined with the Xiaoice deep neural network rendering technology (XNR), the overall naturalness including faces, expressions, body movements, etc. can be greatly improved.
Finally, learning the technique through a small sample of Xiaoice framework, which only takes a one-week training cycle, these two digital twins are born.
Sounds relaxed?
In fact, from the training cycle alone, the training cycle of virtual people in the past was several months, which also caused the problem of long and high cost of virtual people. But after virtual people have become a global hotspot again, this problem must always be solved.
Academia and industry have already begun to act. In November this year, the cover topic of the "Communications of the Chinese Computer Society" was "The Integration of Computer Graphics and Artificial Intelligence", which discussed the extensive integration of CG and AI from technology to application with 6 professional articles, and hoped to introduce computer vision, natural language processing, computer speech and deep learning methods into CG, hoping to transform human-driven into data-driven, upgrade manual algorithms to automatic algorithms based on data analysis, and improve the overall ease of use and efficiency of the system.
The specific digital twins N Xiao Black and N Xiao Bai mean that a long-standing problem that plagues the entertainment and media industry – that is, "how to render virtual people efficiently and realistically" – may have a new solution. Netizens who are not afraid of revealing their age should remember the brave attempt of "Final Fantasy VII: The Arrival of the Son" 16 years ago, and the anti-heavenly details of "Avatar" - this is more than a decade ago, when the film and television industry produced virtual people The time cycle and cost are in years and millions of dollars (still just the protagonist's hair rendering).
Why Xiaoice solved this problem with deep neural network rendering technology (XNR) fused with CG? On the one hand, compared with traditional CG, this technology can make full use of deep learning, extract data a priori from large amounts of data, and achieve acceleration and lightweighting of certain modules in the traditional rendering process.
Simply put, it is to swim faster.
On the other hand, the three-dimensional scene expression based on deep learning and the corresponding neural rendering pipeline can play a role in specific scenes, making the visual image, expression and body movement of the virtual person more realistic.
Simply put, it is better to do so.
Solving this problem, this technology can create great business value.
However, it is not enough to use CG and AI to integrate to create people, how to present virtual people, what content to present, for the general public, is the most important.
And the biggest difference between N Xiao Black, N Xiao Bai and other virtual anchors is also that -
Their image is not only generated by AI modeling, from the text generation of financial information, to the synchronous broadcast of sound and picture through pre-trained virtual anchors, combined with the dynamic drawing of synchronous scenes, and finally generating a complete live broadcast push stream, all of which are completed by AI and do not require anyone to participate.
This is also the first time in the world that the whole process of AI acquisition, editing and broadcasting has been truly automated.
In this regard, in the view of Li Di, CEO of Xiaoice, "an era of AI Being that never tires, is safe and reliable, and has stable output has arrived."
N small black and N small white, what kind of future does it indicate?
Now, we are becoming more and more familiar with virtual people.
The process of virtual digital technology behind them is fast.
In terms of image anthropomorphism, Luo Tianyi and Ling Yi set off a boom in two-dimensional virtual people, and now the hyper-realistic virtual idols such as Ling _Ling, Ayayi, and Ah Xi are also getting hotter and hotter, and then looking back, the future of digital twins is getting closer and closer to us.
From Huang Jenxun "cheating" everyone at the NVIDIA press conference to the latest N small black N small white Xiaoice, it can be seen that all kinds of virtual people have gone wider and wider in the pursuit of the "road of true nature".
On the road of real nature, there are bigger obstacles, such as virtual human content, interaction, diversity, etc.
Traditionally, the industry usually uses a combination of CG and motion capture to create virtual people, completely relying on manpower, long cycle, high cost and not to mention, the key is no content, no interaction, and fixed human design.
From this point of view, N Xiaohei and N Xiaobai are creating a full-capability picture of the framework after the fusion of CG and artificial intelligence through the artificial intelligence technology of Xiaoice.
For example, from the perspective of content and interaction methods, the traditional virtual person will either only speak, sing and interact after the plane hard photo (face change), or the text (manual operation), plus the middle person.
And the digital twins N Xiao black and N Xiao Bai, who can already express themselves freely visually, sound, and text, also indicate that the interaction of virtual people is about to achieve multimodality - remember Rinna who took everyone to the Japanese aquarium two years ago?
It can be said that the emergence of two digital twins, N Xiao black and N xiao bai, is expected to open a new future for the popularization of virtual people in various scenes. Xiaoice team also said that more virtual people will be launched in the near future, and from the perspective of official announcement materials, Xia Yubing, whose works are participating in the China Pavilion of the Dubai World Expo, is also about to open.