Looking forward to it, looking forward to it, the footsteps of spring are approaching, and Xue Rongrong has finally officially taken up his post and ushered in his home stadium - the 2022 Beijing Winter Paralympic Games.
This Winter Olympics is not only a competitive stage for athletes, but also an excellent stage for technology companies to "show muscles". Among the many technological highlights, you may have noticed that the ice pier and the snow melt have a common "colleague" - sign language digital people.
At the Winter Olympics that opened on the evening of February 4, the CCTV News AI sign language anchor officially took up his post and accompanied the hearing impaired to witness a wonderful game. At the upcoming Winter Paralympic Games, the sign language digital man is also obligated to wear a dress to let the hearing impaired feel the charm of ice and snow sports in real time.
Creating a rich legacy of the Winter Olympics, bringing long-term and positive benefits to the country, the host city and the people, is also one of the important symbols of successful Olympic games. In daily life, can sign language digital people continue to shine and provide services for the hearing impaired?
We are concerned that on March 3, Baidu Intelligent Yunxi, the producer of the AI sign language anchor of CCTV News, launched the "AI Sign Language Platform", which proposed a new solution for the popularization of sign language services through the ability to generate sign language synthesis videos at the minute level and live broadcast of sign language anchors in real time.
At the same time, Baidu Intelligent Yunxiling also released the "AI Sign Language Platform All-in-One Machine", so that some scenarios that require hardware interaction, such as hospitals, banks, stations and other public places, plug in to provide sign language services and quickly deploy barrier-free windows.
The innovation of platformization and soft and hard collaboration is putting sign language digital people on an evolutionary path that has long-term docking with social values and grown together.
Tech giants are actively building sign language digital people, reflecting the direction of the tide? What does it mean that the warmth of digital life and intelligent technology accelerates into reality?
When Digital Life Awakens: Sign Language Digital Human Capacity System
What special abilities does Baidu Intelligent YunxiLing Platform give to sign language digital people? Let's look at it in terms of the standards of a human sign language teacher.
There is a kind of "difficulty", called Zhu Guangquan's sign language teacher, who wants to translate Zhu Guangquan's witty words in real time and accurately, and the thousands of CCTV sign language teachers are sometimes busy. In the previous online pk with Zhu Guangquan, the first AI sign language anchor created by "Baidu Intelligent Yunxi Ling", in the face of Zhu Guangquan's continuous throwing of ultra-high-speed smooth mouth, can immediately react, showing smooth and accurate business capabilities.
On the whole, the gorgeous skills of sign language anchors and the solid services at the Winter Olympics come from the basic capabilities provided by Baidu's intelligent Cloud Xiling platform:
1. Comprehension ability.
In the real world, it is easy to be disturbed by noise, and human sign language teachers must hear and understand the news content clearly, otherwise the translation may be wrong, and the comparison is as fierce as a tiger, but it cannot really be used.
To hear clearly, you need leading speech recognition capabilities. Baidu Intelligent Yunxiling Platform integrates Baidu natural language processing technology, mature and leading full-duplex ASR (Automatic Speech Recognition) speech recognition model, and the recognition accuracy of near-field Chinese Mandarin can reach more than 98%.
Easily get all kinds of voice content, even the divine speed of the paragraph hand Zhu Guangquan is not under the words, which lays a solid foundation for the subsequent digital person's sign language translation, making the AI sign language platform all-in-one machine better applied to different scenarios.
2. Translation ability.
In addition to perception, the sign language teacher analyzes and summarizes important information, refines and adjusts the word order according to the overall meaning of the sentence, and converts it into a sign language.
Some manufacturers develop sign language digital people directly use "gesture Chinese corpus", the advantage is that there is no need to re-label, saving time, the problem is to rigidly connect sign language gestures in accordance with the order of speaking, and can not be regarded as "human high-quality sign language".
For example, "I want to go home" is not to compare these four Chinese characters in turn, but to express them in the order of "home", "back", and "I think".
Therefore, in order to translate accurately, sign language numeral people must learn natural sign language word order. Based on the "National Sign Language Grammar Rules", Baidu Intelligent Yunxiling Platform has invited hundreds of hearing impaired students to do data annotation in conjunction with sign language linguistics experts, special education experts, Tianjin University of Technology, etc., forming nearly 10 million high-quality training data.
With the data, the next step is model setup and training. Based on Baidu's neural network translation technology accumulated over the years, the translation method from Chinese text to sign language symbols has been designed, and the industry's first controllable sign language translation model based on neural network has been created, so that the translation comprehensibility of sign language digital people has reached more than 85%, which is comparable to the mainstream machine translation results in Chinese, English, China and Japan.
3. Expressive ability.
In sign language, gestures are essential, and body language such as expressions, mouth shapes, and movements is also required to help people with hearing impairment better understand. For example, the question sentence "Have you eaten", not only to make the gesture of eating, but also to match the expression of doubt, the brow is wrinkled, and the eyes are wide.
It is not a small technical difficulty to make sign language digital people express themselves in a harmonious and dancing way, especially 3D portraits. Some sign language digital people move too fast, and sometimes there are incoherent cases of stuttering. In order to train the "sound table" of sign language digital people, Baidu Intelligent Yunxi Ling Platform is also painstaking:
In terms of expressions, Baidu Intelligent Cloud has accumulated more than 10,000 facial 4D data with 4D scanning data, and with the help of high-precision digital people's "text-to-shape cross-modal facial expression generation technology", it can accurately generate smiles, happy smiles, winks, bubble blowing, white eyes, thinking and other expressions. The accuracy of lip sync is 98.5%, and letters with close expressions when pronounced a and e can be carefully distinguished.
When driving, through personalized TTS, adaptive according to the input text/voice information, combined with a variety of preset actions, drive the digital person's lip shape, limbs, expressions, gestures and other automatic generation. Multimodal sign language expressions that convey richer, more accurate, and easier to understand information.
At the same time, the open domain dialogue platform PLATO-XL equipped with Baidu's intelligent Cloud Xiling platform is trained by Baidu based on tens of billions of training parameters, years of search and knowledge graph accumulation, and is considered to be the largest Chinese-English dialogue model at present. Through it, it can quickly drive digital people to achieve live broadcasts, animations and other content, and achieve real-time communication in multiple scenarios.
Looking at the ability system of sign language digital people, it is not difficult to find that the head technology companies have successively launched their own sign language robots, in addition to reflecting the humanistic care of science and technology, but also hiding the inevitable development of technology.
We must have strong capabilities in computing power, data, and algorithms, and have a leading edge in speech, vision, NLP, knowledge graphs and other fields, so that sign language digital people can truly awaken in front of the screen and live.
As a company with a more complete layout of AI technology in China, Baidu can realize the large-scale application of sign language digital people as quickly as possible, which is why.
Platform-based replication of digital life: Sign language digital people plug into the wings of the industry
The large-scale application at the Winter Olympics and the upcoming Winter Paralympic Games represents almost the highest level of digital virtual people at this stage, and is typical of digital life: the ability to accomplish complex goals (to deliver event information through sign language translation) and the ability to learn evolution in real time (to gather information, interact in real time, respond, rather than record in advance).
As Max Tegmack, the founder of the Future of Life Institute, put it, digital life is a self-replicating information processing system, with physical structures being its hardware and behavior and "algorithms" being its software. This determines that sign language digital people must develop in the direction of soft and hard collaboration and scale replication.
On March 3, Baidu Intelligent Yunxiling released the AI sign language platform and the "AI sign language platform all-in-one machine", which may be inserting the wings of the industry for sign language digital people.
Why? Although sign language digital people are good, they cannot underestimate the difficulty of technology industrialization, and there are at least a few mountains in front of them:
The first mountain is the mystery of efficiency.
For the emerging field of sign language digital people, the production is difficult, the cycle is long, the technical threshold is high, the service group is relatively small, many industries and enterprises will worry before the introduction, will not require a lot of human and financial costs, will not be bad for no one to use, think before and after that is, wait and say. To allow the whole society to enjoy the technology dividend, or to respect the laws of the industry, reduce the application threshold of new technologies, and truly let the production of sign language digital people "reduce costs and increase efficiency".
Baidu's intelligent Yunxiling sign language digital human platform appeared at the right time. The "AI Sign Language Platform" has four major functions: "Video Sign Language Synthesis", "Live Sign Language Synthesis", "Text to Sign Language" and "Speech to Sign Language", which can realize the integration of ordinary video into sign language video, the addition of sign language screens in real-time live broadcasts, the translation of text into sign language, and the real-time translation of speech into sign language. The AI sign language platform can be installed in various apps, websites, and mini programs, so that the hearing impaired can easily realize various needs such as online social networking, entertainment and leisure, and course learning.
At the same time, Baidu Intelligent YunxiLing has also set up three major platforms so that sign language digital people can be produced and delivered quickly, standardized and efficiently. For example, on the human resources management platform, different personas are set up according to different scenes, such as the sign language digital people introduced in the bank can be professional and rigorous, and the sign language digital people used in scenic spots are friendly and lively, etc., to meet the needs of thousands of industries.
The ability of platformization, standardization and systematization makes the AI-driven 2D digital people only need a few hours in the production cycle, and the 3D virtual idol can be developed in a week or two, easily flying over the mountain of efficiency.
The second mountain is the dilemma of experience.
You may have noticed that before Baidu Intelligent Yunxiling released the "AI Sign Language Platform All-in-One Machine", almost all sign language digital people existed in the form of software. Is it really necessary to create a sign language digital person hardware?
Fundamentally, all life forms we know have a carrier of biological "hardware", and some technicians believe that digital life in the "life 3.0" stage must not only have the evolutionary ability to design its own software, but also design its own hardware.
Many banks, hospitals, etc. are introducing humanoid intelligent robots to increase the user's sense of experience. Specific to sign language digital people, as a future service carrier in the fields of social networking, e-commerce, live broadcasting, customer service, tour guides, etc., the key entrance for enterprises to interact with hearing-impaired users is obviously not convenient enough if they can only interact through software.
However, the development of a humanoid sign language robot involves a rather long and complex industrial chain, which can easily deter enterprises.
Baidu Intelligent YunxiLing released the full offline all-in-one machine V3 and the end cloud combined with the all-in-one P3, equipped with the core function of the "AI Sign Language Platform", AI sign language digital people can be quickly and mass-produced like mobile phones and computers, and serve the hearing impaired in all corners of offline life.
Among them, the local all-offline all-in-one machine, in some areas with poor network conditions, such as remote mountain villages, scenic spots and other places, can still carry out sign language translation, portrait rendering and other operations, providing text to sign language, voice to sign language and other services.
The combination of the end cloud and the all-in-one machine can flexibly implement sign language services through cloud computing + local rendering.
The third mountain is the difficulty of evolution.
One of the criteria for measuring a digital life is the ability to learn independently, adapt independently, and evolve itself, which requires comprehensive AI capability support. At present, the entire industrial chain of sign language robots has not been fully opened, although some companies have played the concept of "sign language digital people", but they can only appear in some occasions and videos.
Promoting the continuous upgrading of sign language digital people in real industrial scenarios is an indispensable ability in the industrialization of AI. Among China's AI technology companies, there are few companies like Baidu that have full-stack AI capabilities from the underlying computing power, development framework to industrial solutions.
At present, Baidu's full-stack AI capabilities are integrated into Baidu's intelligent cloud Xiling, which brings unlimited potential to the upgrade ability of digital people, and will also accelerate the full scene coverage of the "new species" of sign language digital people.
Through deep integration with the industry, sign language digital people will also become more and more complex and intelligent, evolving into a real digital life.
At present, there are nearly 27.8 million hearing impaired people in the mainland, while there are only about 10,000 sign language interpreter teachers, and many scenarios cannot quickly keep up with sign language services, which can easily cause new injustices in a rapidly developing and changing society.
The AI sign language platform of Baidu Intelligent YunxiLing makes the large-scale replication of sign language digital people more feasible; the "AI sign language platform all-in-one machine" makes the experience brought by digital person technology richer and more diverse.
The platform replication of digital life is the premise that social responsibility will not become empty talk, which means a reconciliation between commercial value and technological inclusion, and also indicates the rapid opening of the digital human market in sign language.
The Invisible Change in the AI Industry: The Chain Reaction of Sign Language Digital People
Platformization and soft and hardware integration, scale replication of industrial landing efficiency and intuitive experience value, so that Baidu intelligent Yunxi Ling in the sign language digital people competition, has gained a first-mover advantage.
In addition to benefiting the disabled population, technological accessibility will also bring unexpected gains to the enterprise itself and the entire industry. The popularity of sign language digital people, in exchange for the expansion of AI audiences and the extension of sign language services, will make many scenes that we are accustomed to appear obvious expansion and innovation, and trigger a series of chain reactions.
First of all, AI sign language solutions continue to replicate in all walks of life, so that people with hearing impairments are happy to use, enterprises and institutions are happy to introduce sign language services, so that the landing scenarios of sign language digital people will become more and more abundant, and the value in the fields of public welfare attributes, social networking, communication and marketing will be revealed one by one.
Secondly, as a recognized entrance to the virtual world, digital people bring huge commercial space and have become the focus of competition for Internet technology companies. Seizing the opportunity of large-scale production of digital people and cultivating the trust and loyalty of the B-end market will help to occupy an advantage in the next market competition.
Further, Baidu Intelligent Yunxiling's leading and comprehensive nature in technology determines its qualification to participate in or even dominate the establishment of digital human industry standards, which will attract a large number of developers and upstream and downstream of the industrial chain to accelerate the convergence into the ecosystem, promote continuous iteration of technology and continuous innovation of applications, explore the business model of digital people in advance, and drive the growth of cloud computing, AIoT and other fields, which plays an important role in the development of China's digital economy.
In the past, when intelligent technology was mentioned, everyone may pay more attention to macro concepts such as unicorns, investment and financing, and digital economy, but now, smart new species such as sign language digital people are bringing convenience to the "small things" in the daily lives of disabled people.
From the winter Olympics anchor to the AI sign language platform, Baidu Intelligent Yunxi Ling proved to the world that only need to open a channel and connect a bridge, and the technology dividend can be continuously gathered among those who need it.
Allowing mankind a better future is perhaps the warmest chapter in the AI story.