In 1982, Lin Mingmei, the heroine of the animated work “Fortress beyond time and space”, became popular with her moving song and pure human design, thus becoming the world’s first virtual idol.
In 2007, the first tone future was born with the blessing of VOCALOID speech synthesis technology, and the heat continues to this day.
In 2021, master tiktok Liu Yexi got about 3000000 points of praise and released millions of dollars after the first short video was released.
At the New Year Party of Jiangsu Satellite TV in 2022, the virtual human based on Deng Lijun and the live singer Zhou Shen sang on the same stage across time and space.
\u3000\u3000……
Virtual human is not the product of the new era. In the hot meta universe, there are more ways to open it.
Under the new era background, what new characteristics have appeared in virtual human? What underlying technologies are needed? What industrial chain investment opportunities will it bring? This issue of hard core investment research interviewed a number of industry insiders to try to answer the above questions.
evolution history of virtual human
According to the qubit deep industry report of virtual digital human, virtual digital human refers to the comprehensive product that exists in the non physical world, is created and used by computer means such as computer graphics, graphics rendering, motion capture, deep learning and speech synthesis, and has multiple human characteristics (appearance characteristics, human performance ability, human interaction ability, etc.). On the market, it is also called virtual image, virtual human, digital human, etc. representative subdivision applications include virtual assistant, virtual customer service, virtual idol / anchor, etc.
Luan Qing, vice president of mobile intelligence business group of Shangtang Technology (00020. HK), said in an interview with 21st Century Capital Research Institute, “virtual digital human is built based on vision, voice, NLP and other technologies, which can simulate real people’s conversation, expression and action, and carry out interactive applications in various scenes.”
\u3000\u3000 “Behind the normal operation of the virtual human model is the continuous support of data and technology. On the one hand, it must continuously obtain high-quality sample data for training. On the other hand, it must also combine the empowerment of technology, such as speech synthesis, multi-modal interaction, deep neural network rendering, etc., so as to make the whole including face, expression, sound, limb movement, etc The degree of naturalness can be close to the level of real people. ” The relevant person in charge of Beijing Haitian Ruisheng Science Technology Ltd(688787) (688787. SH) told 21st Century Capital Research Institute.
It is not difficult to find that an important feature of virtual human is that it can simulate real people and interact. Many insiders told the 21st century capital research institute that this means that virtual people “can move themselves”, which is different from NPC in traditional games.
Take the reproducing singer Teresa Teng as an example. In 2013, with the help of the technical scheme provided by the digital Kingdom, the visual effect team of “rejuvenation” allowed Teresa Teng to appear at the little big egg site of Jay Chou’s “magic Tianlun” World Tour Concert Taipei station and sing in duel with Jay Chou. At that time, the attention of the market was limited to the appreciation of technology.
Ten years later, Teresa Teng received varying degrees of attention and market feedback. In this new year’s Gala of Jiangsu Satellite TV, the digital Kingdom enables virtual Deng Lijun to interact with people in real time through technology, which means that more complex actions are captured and rendered in real time, and the time cost is relatively high.
Chang Fei, product director of Wanxing Luyan, a video demonstration product of Wondershare Technology Group Co.Ltd(300624) (300624. SZ), analyzed the 21st Century Capital Research Institute, “With the continuous development of technology, virtual people have experienced early manual drawing, computer drawing and artificial intelligence synthesis. Virtual people are gradually simplified. At the same time, based on the application expansion of artificial intelligence technologies such as natural language processing, speech recognition and planning vision, virtual digital people are developing towards intelligence, convenience, refinement and diversification. At present, virtual people are in appearance, behavior and communication Highly anthropomorphic in all aspects. “
dismantling the virtual human industry chain
Combined with the calculation of qubit, if the industrial application is successfully implemented, the market scale of Chinese virtual human will reach 270 billion yuan in 2030, of which the contribution of identity virtual human will exceed 170 billion yuan. In the process of development and upgrading of the whole industrial chain, high-quality investment opportunities will continue to emerge. Compared with those too distant concepts of the meta universe, “virtual man” seems to have become a track within reach at present.
“With the reasons for the epidemic in the past two years, video has become a better way of communication and expression. Some people, such as some teachers, we media and enterprises, are reluctant to show real people when making videos, which leads to the trend of using virtual images instead.” Changfei said.
At present, head Internet companies have entered the bureau with virtual people as the entry point to increase capital investment.
Tencent, Baidu, Alibaba, Netease, Baidu and other Internet companies all invest in virtual people; Dayu network, Cishi culture and other MCN companies have broadened the design and operation business of virtual image; Bluefocus Intelligent Communications Group Co.Ltd(300058) and other marketing companies continue to strengthen the marketing service ability in the field of virtual human; Small red book, jitter and other social platforms tiktok and drain the virtual idol bloggers.
From the perspective of industrial chain, virtual digital people can be divided into upper, middle and lower links. The upstream industrial chain includes companies that produce content, tools and IP planning, such as Microsoft, Houdini, Autodesk, apple, readtext group, etc. Before the birth of virtual human, it needs content production and IP planning to determine its character and image. The infrastructure also includes hardware manufacturers such as display equipment, optics, sensors and chips, as well as software manufacturing such as modeling software and rendering engine. In the later stage, it needs technical support such as modeling binding, driving and rendering.
The midstream industrial chain is mainly virtual digital human manufacturers, including software and hardware systems, production technology service platforms and AI capability platforms, including various enterprises providing voice recognition, CG modeling, XR and other technologies, such as Iflytek Co.Ltd(002230) (002230. SZ), avatarworks, Tencent, Xiangxin technology, Huoshan engine, Baidu, etc.
For example, Dalian Zeus Entertainment Co.Ltd(002354) (002354. SZ) recently said on the investor interaction platform that the company has recently established a new holding subsidiary, Beijing Yuanjing Digital Technology Co., Ltd., whose main business is to build a virtual digital human production platform, develop virtual digital human such as virtual anchor and virtual idol, and serve E-sports games, brand marketing and other fields.
The downstream industry chain includes media, games, film and television, finance, culture and tourism, education, medical treatment and other fields, such as various virtual hosts, virtual anchors, virtual idols, intelligent customer service, intelligent financial consultants, virtual tour guides, commentators, etc., forming an overall industry solution and enabling the development of various fields.
Beijing Jetsen Technology Co.Ltd(300182) (300182. SZ) recently said that its company has officially launched the virtual human “Miao Jiangshan” recently, and plans to take the lead in trying the commercial realization model in the fields of commercial endorsement, live broadcast, short video and so on
In addition to the above industrial chain related companies, Citic Securities Company Limited(600030) also suggests paying attention to roblox and other platform companies that combine content IP operation and R & D capabilities; Tiktok Kwai, bubble Matt, Mango Excellent Media Co.Ltd(300413) (300413.SZ), Col Digital Publishing Group Co.Ltd(300364) (300364.SZ) and other content companies with rich digital IP resources and excellent operational capabilities, as well as voice, fast, little red book, micro-blog and other virtual human content operation platform company.
In addition, Shenzhen Mason Technologies Co.Ltd(002654) (002654. SZ), Hangzhou Anysoft Information Technology Co.Ltd(300571) (300571. SZ), Zhejiang Jinke Tom Culture Industry Co.Ltd(300459) (300459. SZ) are also concerned or laid out.
However, it is not difficult to see that there are no pure virtual human targets in the current A-share market, mostly providers of specific technologies in a certain industrial chain link.
underlying technical support opportunities
The new demand for virtual people in the current market needs a series of technical support. This has also led to many investment opportunities.
Relevant researchers of wisdom bud said that as an emerging comprehensive technology application field, virtual human mainly involves technical fields such as graphics rendering, motion capture, speech recognition, natural language processing, multimodal technology and deep learning.
Taking the upstream as an example, according to the search of smart bud global patent database, Microsoft and its affiliated companies have more than 3000 patent applications applicable to the field of virtual human, mainly focusing on the fields of speech recognition, natural language processing, deep learning, computer vision and so on. The technical layout of Houdini (side effects software) in this field mainly focuses on computer graphics, animation and other fields.
Yuewen group is a comprehensive cultural industry group based on digital reading and centered on IP cultivation and development. The group and its affiliated companies have no patent applications directly related to the virtual human field.
In the midstream link, smart bud data shows that the technical layout of Iflytek Co.Ltd(002230) and its affiliated companies in this field mainly focuses on the technical fields of speech recognition, speech synthesis, knowledge atlas, image recognition and so on. The technical layout of Baidu and its affiliated companies mainly focuses on the fields of in-depth learning, computer vision, natural language processing, image processing and so on.
Downstream, Netease’s technical layout in this field mainly focuses on virtual roles, touch operations, computer graphics and other fields. BiliBili’s technical layout in this field mainly focuses on virtual image, image rendering, speech recognition and other fields.
Changfei further analyzed that the underlying technology of virtual human includes 3D image design and modeling, model binding, motion capture and driving technology of face, half body and whole body, 3D rendering technology, etc. At present, the barriers of motion capture, driving and 3D rendering technology are relatively high. Virtual people are divided into cartoon 3D virtual people and high simulation virtual people. In particular, the latter has high computational complexity for motion refinement and the whole technical process, and high requirements for real-time algorithm effect, so it is difficult to do a good job.
It is reported that Wanxing has rich experience in 3D image design and modeling. At the same time, the AI / 3D / Ar / VR technology team led by the doctor has been able to realize the 3D cartoon of real-time video avatar, the motion capture technology of face and body, and the 3D rendering ability of application end.
Shangtang technology, as a technology provider, is in the middle and upper reaches of the industrial chain. Luan Qing told the 21st Century Capital Research Institute, “the technical difficulty of digital people lies in the accurate expression, fluency and naturalness of expressions and movements. The difficulty in landing lies in adapting to different application scenarios.”
It is reported that Shangtang’s sensemars agent digital people have been trained in knowledge databases in different fields and have been applied to shopping centers, banks, online customer service, museums, exhibition halls, scenic spots, airports and other industries and fields. For example, in terms of interaction, based on the AI technology of Shangtang, sensemars agent digital human can realize interactive applications such as accurate mouth shape, realistic action, and intelligent dialogue with real people.
A-share listed company Beijing Haitian Ruisheng Science Technology Ltd(688787) is a provider of artificial intelligence data and related data services.
The relevant person in charge said, “the current application of virtual human is relatively cutting-edge, and its final state can be realized only with the support of characteristic, diversified and high matching training data, that is, better data can train more realistic virtual human.”
Obviously, to a certain extent, the virtual human track, which is regarded as the infrastructure of the meta universe, has left a lot of imagination for the outside world.