[TechWeb] Among the first batch of artificial intelligence enterprises listed in China, Keda News is unique, focusing on the voice track and maintaining a long-term lead.
In 2017, HKUST's net profit for the first half of the year was 107 million yuan, down 58.1% year-on-year; in 2018, the annual revenue was 7.917 billion yuan, up 45.41% year-on-year; the annual net profit was 542 million yuan, up 24.71% year-on-year.
During the 2019 Suzhou Zhibo Fair, Li Shipeng, vice president of Keda Xunfei, accepted an interview with TechWeb and discussed the development and application direction of Keda Xunfei products and the market competition.
"The voice field has no rivals"
TechWeb: In the past two years, many start-up voice solution providers have entered the market in Zhongguancun and Shenzhen. How do you see the influx of such competitors? Does HKUST have enough technical barriers to deal with it?
Li Shipeng: If you go to the CES show, more than 80% of the Chinese conversation products are based on the technology of the University of Science and Technology. Including Haier's big factory, we also made a joint venture with Haier not long ago. You can also see that we can work with large car manufacturers to do in-vehicle voice control.
To put it bluntly, iFLYTEK and the small new entrants are not at the same level, they are not the real competitors of iFLYTEK.
IFLYTEK technology has three major advantages, can get rid of "competitors": one is in the noisy environment function pick-up performance; second, dialect recognition accuracy, iFLYTEK supports 24 dialects, no second company can do so at present, the second is the ability to pick up sounds in noisy environments, the second is the accuracy of dialect recognition, iFLYTEK supports 24 dialects, and no second company can do so at present. In addition, under different Domain (domain), the translation of professional vocabulary is very different, and a word has different meanings in the field of medicine and IT. In these aspects, we have run ahead of most enterprises.
"iFLYTEK's OCR is very strong, but it never propagandizes."
Li Shipeng said the iFLYTEK Institute of artificial Intelligence, which is familiar with the outside world, focuses on speech-related technologies, such as speech synthesis, speech recognition, dialect recognition, noise removal, and so on. But at the same time, in addition to voice, iFLYTEK is also doing deep computer vision. Even if the media generally think of Shang Tang, open vision and focus on vision, but at least in the field of ORC, we believe that there is no second domestic competition with iFLYTEK.
TechWeb notes that Li Shipeng used the word "second to none" in describing iFLYTEK ORC technology.
IFLYTEK attaches importance to intelligent medical treatment. The technique emphasizes understanding the patient's statement and requires machine understanding before giving advice. In terms of natural language understanding, iFLYTEK achieved the first time that the accuracy of machine English understanding exceeded that of human beings in the Stanford SQuAD Reading comprehension Competition in recent years.
Deep learning is just a framework for solving problems. It is still the data that really opens up the gap between enterprises.
"Useless research and development has no room for survival"
TechWeb: How does CoD News coordinate the relationship between technology R&D and commercial landing? Does Xunfei Research Institute have medium and long-term research and development projects?
Li Shipeng: First of all, there is no room for survival in research and development that has nothing to do with actual needs.
IFLYTEK invests more in research and development than any other company. It can be said that in addition to R & D or R & D. Most of our senior leaders graduated from the Chinese University of Science and Technology. Chinese college students have an endless love of technology, so Yu Li is not required to spend all kinds of technology.
We have a set of internal R & D to product landing process. The iFLYTEK Institute only does useful research, focusing on technologies that can be realized in 1-3 years and 3 years. IFLYTEK will not invest too much resources in the pursuit of unrealistic projects.
In order to maintain the leading position of technology, our main personnel invested in research and development in 3-5 years. If the product encounters difficulties, the institute will focus on a number of “study teams” to promote.
When time matures, iFLYTEK will have a steady stream of short, medium and long-term technological reserves, while useless research projects have no room for survival.
"AI landing is like finding a nail with a hammer"
With regard to the landing of AI, Li Shipeng believes that there are currently two types of AI enterprises on the market, one is technology companies, such as "looking for nails with a hammer", hoping to apply the existing technology to suitable areas. The second is that the traditional enterprises of AI transformation, such as Haikang and Huawei, have their own strong support and take a road to bring AI into traditional products. Each of these two types of companies has its own strengths.
Keda Xunfei belongs to the former, playing the strengths of a technology company in the field of voice, while opening up new industries.
When Li Shipeng talked about the cooperative relationship between the two types of enterprises, he said: the cross-industry application of technology companies must be a cooperative relationship. A video surveillance company, for example, wants to work with someone's face recognition company, which favours the accumulation of the former's product market and the latter's AI technology breakthrough.
At first, each of the two will lose both odds of grabbing market share. Only by cooperation can we give full play to their respective advantages. Slowly, the two technology levels are not much different, so we want to do two on their own, so we see that traditional industry companies are starting to set up AI Research Institute. This is also the inevitable situation of AI industry landing.
The AI landing idea of technology companies is first of all to consider how to survive. First, artificial intelligence still has to work with traditional industries. Only the concept and separated from the specific industry, technology can not land. For example, human face recognition, ideal recognition accuracy is very high, but to consider the actual situation such as ambient light, object interference and so on. The general situation in the industry is that the ideal environmental data is very good-looking, but can not meet the application standards. So we have to return to cooperation.
Second, technology companies in the landing while exploring their own new industry. IFLYTEK's translation machine (To C) product line belongs to the new industry. There were no such products on the market before, but iFLYTEK can integrate offline translation and other functions into the translation machine, and now some translation products have gradually become a necessity for specific groups.
(text / walker Tao)