Voice of Innovation
Jiang Tao, Senior Vice President and co-founder of iFlytekNI YANSHUOWhen Chinese Premier Li Keqiang said, “Hello, Prime Minister Borissov,” in Chinese to a small machine, his words were immediately translated into Bulgarian. Premier Li was introducing an artifi cial intelligence (AI) translating machine developed by a Chinese enterprise to the leaders of Central and Eastern European countries while attending an exhibition on the results of local cooperation between China and the 16 Central and Eastern European countries on July 7.
Bulgarian Prime Minister Borissov was very impressed by the small translator. Premier Li bought the device on the spot and presented it to Borissov with delight. Borissov then spoke a sentence in Bulgarian and his words were rendered into Chinese in no time: “Thank you, China’s premier!”
The machine developed by iFlytek Co. Ltd. can quickly and accurately translate between Chinese and 33 other languages including English, Japanese, Korean, French, Spanish, German and Russian. It can provide auto-translation services in real time, making daily communication during travel, work and medical diagnosis more convenient. The translation device was released by iFlytek in Beijing on April 20.
Since the company was established in Hefei, capital city of east China’s Anhui Province, in 1999, its market value has increased from 3 million yuan ($449,000) to 80 billion yuan ($11.97 billion). It has become the largest listed intelligent speech and AI company in the Asia-Pacific region and a global leader in several technologies including speech synthesis, speech recognition and natural-language processing. The new device testifi es to iFlytek’s innovative ability and growth.
From newcomer to champion
IFlytek was originally the Man-Machine Speech Communication Laboratory of the University of Science and Technology of China. Supported by the State Hi-Tech Development Plan, or 863 Program, the lab made synthesized voices clear and natural through its independently developed system in 1998.
“I thought we could develop some products, since our technology was already functional,” recalled Jiang Tao, co-founder and Senior Vice President of iFlytek. At the end of 1999, iFlytek was established with 3 million yuan of registered capital. The goal of the company was to make all machines and devices such as mobile phones, computers, televisions, cars and toys able to listen and speak as humans do.
However, the growth of the company was not as smooth as Jiang had imagined. In fact, the company made no profits from 1999 to 2004, and it even failed to pay its employees for some time.
Its fi rst product targeting the consumer market was iFlyBook, a software enabling people to operate computers through voice.
“After the software came out, we found that people had no such demand since they were accustomed to using the keyboard and mouse,” said Jiang. He explained that the household bandwidth was not wide enough, while cloud computing and the mobile Internet system were not fully developed yet, so the product was a failure.
IFlytek made many similar attempts, which were questioned by many stakeholders and the public, but was unable to identify a clear development path until 2001. At that point, the company held a conference in Chaohu, Anhui Province, announcing its goal to develop intelligent speech products.
“In the initial phase, the greatest pressure we faced was how to fi nd a market for our key technologies. The company was on the brink of collapse during that period,” Jiang recalled.
It was at the China Hi-Tech Fair in 2000 held in Shenzhen, south China’s Guangdong Province, that the intelligent speech technology presented by iFlytek attracted the attention of Huawei Technologies Co. Ltd., a leading global information and communications technology solutions provider. Huawei decided to make a bulk purchase of iFlytek’s intelligent speech technology and integrate it into its products.
“That was our fi rst bucket of gold. Hence, we developed our new sales model, which is called iFly Inside, whose purpose is to sell our speech technologies, which are integrated into other products,” said Jiang. In 2004, iFlytek was fi nally able to make ends meet.
Another aspect of iFlytek’s growth came in 2005 from ringtones, which were popular among Chinese consumers. At the time, mobile operators had a problem: How to enable users to fi nd what they wanted from thousands of songs.
IFlytek developed a ringtone system through which users could call in and find the music to download directly by saying the names of songs or singers, or even parts of songs. The system was quickly adopted by China Mobile, China Unicom and China Telecom, the three major telecommunication operators in China. During that period, almost all the speech services related to music and ringtones of these three companies were provided by iFlytek. As a result, the company experienced a 130-percent increase in its net profits for three consecutive years. In 2007, iFlytek’s annual revenue exceeded 200 million yuan ($29.94 million).
IFlytek then began to develop its own products and applications, with its independent product and market development capabilities continuously improving. In 2008, the company was listed on the Shenzhen Stock Exchange.
In recent years, iFlytek has made several breakthroughs in intelligent speech and has become a leader in the industry. It has ranked fi rst for several consecutive years in top global contests on speech synthesis and has achieved the goal of enabling computers to imitate human speech through deep learning methods.
Liu Qingfeng, Chairman of iFlytek, presented former U.S. President Barack Obama with a video produced with the company’s speech synthesis technology in 2017. In the video, Obama gives a speech in English, but also makes a few remarks in perfect Chinese thanks to the technology.
As for speech recognition, iFlytek ranked first in all indicators in the fourth CHiME Speech Separation and Recognition Challenge in 2016, the most influential speech recognition contest in the world. Compared with the 2015 champion, the iFlytek products’ percentage of mistakes declined by 50 percent and broke the world record by a large margin. In 2017, the precision of iFlytek’s online speech recognition platform reached 98 percent. Continuous advances have also been made in its dialect-recognition function covering 22 dialects.
From listener and speaker to thinker and learner
In 2013, iFlytek initiated a SuperBrain Plan, intended to advance the abilities of machines from listening and speaking to thinking and learning.
According to the plan, machines can use deep learning techniques for all rulebased and logical issues relating to human brainwork to solve practical problems. “They can help humans in all aspects,” said Jiang.
In November 2017, iFlytek’s intelligent doctor assistant received a very high score on the National Physician Qualification Examination, becoming the first AI system in the world to pass such an exam. This indicates that the system has been equipped with the professional knowledge and skills required to be a physician.
Doctors at the Anhui Provincial Intelligent Hospital demonstrated how the intelligent doctor assistant can help make diagnoses and provide treatment. After obtaining information about a patient’s condition through several medical exams such as inquiry and medical imaging, an AI doctor assistant will help with diagnosis and give feedback. Doctors can then make a final assessment based on the AI doctor assistant’s suggestions and their own diagnosis, saving considerable time.
“IFlytek’s AI doctor assistant system has been able to help doctors with diagnosis at community-level hospitals,” said Jiang. “This can solve the problem of the uneven distribution of medical resources in China. After all, doctors at top hospitals are still in short supply.”
The company’s achievements in education technology are also quite notable. Currently, machines developed by iFlytek have reached an expert level in terms of checking homework and exam papers. The iFlytek machines can process objective questions and even subjective questions such as essays.
“Machine-based homework checking not only saves teachers’ time but also enables teachers to collect students’ learning data in the process. This helps tailor teaching to students’ individual aptitude, as Confucius pointed out more than 2,000 years ago,” said Jiang. In schools using the machines, the homework for students can be different, which improves students’ learning effi ciency. Students’ average time spent on homework per day has declined by 1.5 hours, Jiang said.
At present, iFlytek’s intelligent education products have been widely applied in 31 provinces, autonomous regions and municipalities, as well as overseas markets including Singapore. The company has cooperated with more than 13,000 schools and served over 80 million teaching staff and students.
In the judicial field, judges usually need to read loads of files before making judgments based on their experience. Now the AI system can assist judges by analyzing files in seconds and providing them with suggestions on three main aspects: charges, prison terms and fi nes.
The intelligent trial system launched by iFlytek has been applied in more than 400 courts and more than 1,000 tribunals across China. The intelligent court products have saved court trial time by 30 percent on average and have enabled trial records to be 100 percent complete, improving the quality and effi ciency of judges’ work.
“The third industrial revolution that took place in the past decades mainly solved the problem of mutual connection, such as the people-to-people connection, the commodity-people connection and the information-people connection. However, it did not solve the problem of information processing after connection. In the decades to come, through AI technology, people can signifi cantly reduce the costs of information processing after connection and improve the productivity of the human brain,” Jiang concluded.
An iFlytek staff member demonstratesan AI application for carsLIU JUNXICopyedited by Rebeca Toledo Comments to [email protected]