Guide: The input method products that are considered by the outside world to optimize the space are getting smaller and smaller, and the opportunity to find a breakthrough from the AI has also provoked various competitions. The AI's intervention may not only be "input", but also the inherent knowledge of the entire "input method."
Most people seem to have a very mature Internet product - the input method, and the incitement continues.
On January 16th, Baidu's input method engaged in a press conference called “AI·New Input Full Sensory Input 2.0”, which announced its declared “the first real AI in China”. The input method "Baidu input method AI exploration version, Baidu senior vice president Wang Haifeng said, "Baidu input method will play an important role in the communication bridge, allowing people to interact into a new phase."
This input method is an input product with a default input mode for full voice input and support for expressions, limbs, etc. for so-called "full sensory" input. It seems that Baidu wants to gain breakthroughs in the input method market by virtue of its consistent AI technology advantages, and the definition of “communication bridge” also indicates that Baidu is trying to explore new gameplay on the word “input”.
In fact, the combination of AI and input method is nothing new, and the “600 million users” jokes (who are being seen) are often hanged by AI technology last year. In the month, the Xunfei input method, which was blessed by the so-called “Xunfei Super Brain Project”, was also launched.
The input method products that are considered by the outside world to optimize the space are getting smaller and smaller, and the opportunity to find a breakthrough from the AI has also provoked the competition for competition. The AI's intervention may not only be "input", but also the inherent knowledge of the entire "input method."
The mature input method market, user demand brings innovation power
AI is deeply applied to the input method, and the market demand is drastically changed, and most of the current product forms are still directly related.
First, the main mobile Internet population has a typical "mixed" demand for language input:
The basic grammar of different Chinese language families is constantly integrated, such as the Northern language system "you go first", When you come to Guangdong and Hong Kong, you will become "you go first". If you are Chinese, you will have a mixed usage of "you should go first";
At the same time, including Sichuan dialect and Changsha dialectThe dialects of the special regions such as the Northeast dialect gradually gained public recognition, and the voices such as jiao (foot) began to be mixed;
part of the trend toward elites, or for people with specific social purposes, Sino-British inclusions became the norm," I don't care", "Emma, I think this idea is really amazing", and so on appears more and more in the input situation.
Secondly, the mobile Internet crowd is further sinking, allowing the input "demand mismatch".
Zhang Xiaolong said that middle-aged and elderly people over the age of 55 occupy a considerable proportion of WeChat users. In fact, mobile Internet products, including WeChat, which require a large amount of input methods, are used in low-tier cities and rural areas. The infiltration of old age is an indisputable fact. On the one hand, these users do not understand the conventional input operation. On the other hand, as a "new netizen", the input demand is more urgent, and the demand has a mismatch. Unfortunately, among these people, dialect habits account for the vast majority.
Finally, all Internet products are reminiscent of the new and new humans after 95s and 00. They gradually occupy the main stage of mobile Internet as time goes by, do not like old things, always pursue Novelty experience.
It is not that the "old product" input method must not be liked by them, but it is easy to form a new competitive advantage by grasping these new human needs.
brings diversity and innovation, AI technology into the input method "香饽饽"
In general, the changed "input" needs do not necessarily need AI to solve However, AI is undoubtedly the most diverse and innovative.
As mentioned in the opening paragraph, Keda Xunfei, who relies on technology to make a decision, is the most active application of input method AI. In June 2018, the version of the input method was introduced to intelligently reduce noise in noisy environments. After active conversion. Can support a variety of dialect input, and can be mixed between Chinese and English. In addition, the so-called "synonymous word intelligent resolution" is also introduced, which can judge "he", "her", "he" and so on.
As a player who has a first-mover advantage in the field of input methods, Sogou does not forget to work hard on AI. CEO Wang Xiaochuan emphasized the AI empowerment input method and search in the year of 2018. It is necessary to "develop voice, dialogue and translation through AI technology". Subsequently, the focus of Sogou AI technology can be directly brought to the camp.On the translation machine hardware, Translator Pro's 42 languages are translated in real time, and speech recognition support is added to the input.
Baidu, which is based on AI technology, naturally will not give up the opportunity to use AI to gain competitive advantage in input methods. However, Baidu's gameplay is somewhat different. The "AI Discovery Edition" released this time simply changed the default input method to voice input. The full sensory input of expressions and limbs is obviously to use AI as the "main business". Not just assistive technology to improve accuracy.
Baidu's approach is closely related to the attributes of AI technology. The original AI is to read human language, recognize human commands and make corresponding feedback. There are many ways for AI to obtain information. They can all be regarded as "input". Users also have a lot of "mobile terminals". Enter "requirements, not just textual information.
For example, the early input method was based on character input (users typed the keyboard, like the code code), and later the voice input became popular (users only need to use the mouth), now, Baidu launched the input context Shooting live (allowing the user to "drive" the other party to make the same expression through their own actions), AR expression (adding a background scene for the user's self-timer), volley handwriting and other AI functions, in fact, is the characteristic of the AI technology itself, in the way of input There is some kind of "dimension" improvement, and users can take care of themselves and follow certain things to "enter".
The meaning of full sensory input is nothing more than the information that the user enters through the “input method”. There are more choices in addition to the text. In this sense, AI is both an input method to enhance product diversity. And innovative technical means, in turn, its own attributes also give the path of the input method to break the old "text information".
As a result, the so-called "AI exploration version" of Baidu input method, regardless of whether the substance is valuable or not, at least gives the model of the AI input method.
In addition to innovation, the input method AI competition also needs to grasp the "details" of user needs
input method is a "small product" that is not too eye-catching but inseparable from the user. In addition to the concept and direction of application of AI technology, the attention to detail in specific operations may be more able to capture users.
1, accuracy improvement is an eternal theme
Just like a mobile phone running score, accuracy is a hard indicator of the ability to test input methods. Up to now, the accuracy of the input method has been the process of “high-level extraction”, and it has become more and more extreme and difficult.
For example, Xunfei likes to promote the accuracy of its voice input, which is said to have increased to 98%, and can input 400 words in 1 minute. Sogou also has similar accuracy promotion. The reason why Baidu dared to launch full voice input, in addition to user needs, is what it calls "leading industry voice technology."
Baidu input method released the "streaming truncated multi-layer attention modeling" (SMLTA) technology for online speech, according to its statement, "the world's first implementation of local attention modeling recognition accuracy exceeds the global Attention model, and "the world's first large-scale deployment of online voice interaction attention model." This technology can solve the "delay" problem of the traditional model, Baidu claimed that it "improved the online speech recognition accuracy by 15%, exceeding the industry's highest level of 15%."
2. On the one hand, the 5G era is coming, and one should pay more attention to the demand of “no network”
brush short video, chat WeChat, play games, shopping... these subway families are standard, In fact, there is another requirement for the input method: how to solve the strong offline identification needs.
The pace of the 5G era is getting closer and closer, but in the environment of trains, elevators, tunnels, and high-density crowds (such as the train station of the Spring Festival), slow and no-net provides more serious challenges for voice input. The more the network is disconnected, the higher the user's demand for the input method is. The frequency of use of the office network is reduced (work and other reasons).
Thus, AI voice offline recognition has recently become a technology for input method products. Sogou claims to be able to translate offline in four languages, while Baidu is focusing on the Deep Peak 2 system for optimization attempts to address the user's multiple input scenarios. At present, Baidu input method can still achieve accurate voice input in offline scenes such as subway, elevator, tunnel or crowded people. The official said that it is “above the industry average of 35%”, and a batch of input methods from the conference impromptu. From the perspective of offline voice recognition PK, Baidu input method is indeed dominant.
3, only one product, but must meet the needs of extreme diversification
As stated in the previous article, in the Chinese environment, the input method has to face diverse input requirements, dialect, English, and different grammars... and the input method cannot be developed separately for each local and individual user groups. Better compatibility with all languages at the same time becomes an important part of the input method "experience".
For example, dialect recognition is already the “standard” of most input methods. However, due to technical limitations, users often need a “entry” dialect switching action. On the one hand, this is as simple as the Internet. Thinking is contrary, on the other hand, it forms a natural barrier to many sinking users who are not familiar with electronic products.
This gives Baidu input method opportunities, its "dialect and dialect, dialect and Mandarin mixed voice input" technology, has achieved seamless switching, users can achieve "free to say." Similarly, Chinese and English inclusions can also be switched back and forth. So far, Baidu input method should be the only input method product that realizes high-precision Chinese-English mixed voice input and dialect-free switching voice input.
After the AI technology PK, the input method and the "new opportunity" you can't imagine
objectively, under the wave of AI technology, the details of the above technological innovations, I believe The major input methods will be filled as soon as possible. After all, the product target is there, and it will be implemented.
However, from this conference, Baidu's input method may still "save one hand". The focus of the AI input method template may not only be AI, but Baidu also intends to subvert the old recognition of the input method. Know that opening new application scenarios and business possibilities for input method applications.
From the Baidu input method AI exploration version of the new interaction, new expression, new exploration of the three major product themes, it tried to combine user needs, Internet product ideas, AI technology trends and other major environments, subvert the input method three Inherent cognition:
1, "Input method is only a C-end product"
The input method is of course used by ordinary users, and it is one of the most successful and popular C-side products. However, the information input of the video dimension may open a new commercial space for the input method.
Baidu input method AI exploration version of the film live, show, the platform needs to provide specific elements. Baidu's overseas input method has been jointly launched with the first independent film "Hornet" of the Super IP "Transformers" series.The Wasp series theme map and AR expression, in addition, with the Hello Kitty, Little Yellow Duck, Little Prince, Greenwich and other more than 100 well-known IP around the world to launch IP skin and expression.
based on this On the other hand, Baidu's input method has also designed customized communication for local users, such as local network red cooperation promotion, and this may be the monetization path that Sogou input method and other products have always dreamed of, with the possibility of B-side cooperation.[123 2, "Input method is the tool product"
throw it when it is used up, call it to go, the input method is the most popular "tool product" is nothing, but the full sensory input The input method has been torn open to meet the needs of the current mobile Internet content. Not long ago, WeChat pushed the small video function of the circle of friends, users can instantly capture video sharing. Obviously, social + pan entertainment (especially short video) has become a potential area for all families to compete for, headlines and WeChat are trying to penetrate into each other's field. and "all sensory input" is a clever place, like AR Applications, In a one-on-one chat and group chat environment, users can directly send content similar to small videos for sharing, and achieve the purpose of small video socialization. This is the social and video fusion that most people are pursuing. At present, the AR application is only in the stage of enhancing the input experience. The social pan-entertainment product is only the imagination space of the future. 3, "Input method can only be used for PC, mobile phone input"
[123 The earliest Chinese character input also requires special "Hanka" hardware, and later into the PC era began to enter the software, the mobile Internet era has brought a variety of convenient input experience.
Now, this process is still Continue, the input method may be different from the PC, mobile phone.
Baidu input method AI exploration version of the "new exploration" launched a volley handwriting, using the finger in the camera within the scope of the sky handwriting, input method You can recognize and generate text. At first glance, this is a novelty game that uses a mobile phone camera and AI technology. In fact, this means that the input method has the opportunity to leave the PC and the mobile phone, for example, in the existing case. Baidu The method has been finding volley handwriting watches, small genius applied on a watch phone.
ThisThis means that volley handwriting or similar AI technology, it is possible to import the input method into more intelligent terminals, open a completely different scene for the input method, and promote the experience update of the intelligent hardware.
Wait and see.This article was written by the author of the cutting-edge technology. The views represent only the author and do not represent the OFweek position. If you have any infringement or other problems, please contact us.