Baidu gives Japanese keyboard app AI dictation technology
Internet giant to keep exporting cutting-edge services to Japan
TSUBASA SURUGA, Nikkei staff writer
TOKYO -- Although it withdrew from Japan's search engine market in 2015 -- unable to crack the Yahoo-Google duopoly -- Baidu never really left the country. It just moved into the artificial intelligence arena.
On Wednesday, Baidu Japan officially launched a voice function for Simeji, its popular Japanese keyboard app. Using AI-powered speech recognition, the third-party keyboard "types out" spoken words and automatically adds punctuation -- a thoroughness that leaves competitors like Google behind.
Baidu founder Robin Li earlier this year said his company is accelerating its shift from an internet company to an AI company.
"For Baidu Japan, this is the first year of the AI era," said Charles Zhang, Baidu Japan's president and CEO, during a news conference here. Zhang added that the Japanese subsidiary will keep importing technology from its Chinese parent, which sees AI as the next big growth area, one that Baidu is heavily investing in.
The shift comes amid signs of slowing growth in Baidu's core search business -- a trend that accelerated last year after a scandal in which internet searches were turning up questionable results.
Simeji's new talk-don't-type function was developed by a special team at Baidu's headquarters. In addition to spelling out words, the technology matches spoken words with relevant emoticons, or kaomoji in Japanese. Baidu has a library of over 50,000 kaomoji.
Baidu Japan acquired Simeji in 2011. The app was developed by two Japanese coders in 2009 for Android users. Since 2014, the keyboard app has been available on iOS and proved popular among teenage and young female users. It has been downloaded around 27 million times.
"The accuracy of our voice recognition in Chinese has reached 97%, overtaking competitors like Google," said Li Chao, a speech architect and member of the development team. Japanese -- with its many heteronyms, words that are spelled the same but pronounced and defined differently -- is particularly challenging for speech recognition software. Simeji's accuracy in the language is at 90%, Li added.
In February, Baidu Japan unveiled an image recognition engine that was developed together with Chinese researchers at the parent company. The service targets news sites, recommending related content and stories based on the images their users have looked at.
Zhang said more AI-powered services are on their way to Japan. One of them, an advertising service, will target Chinese tourists. Japan in recent years has been attracting record numbers of foreign tourists, many of whom come from China.
Baidu Japan faced its own scandal in 2013 having to do with Simeji, which was found to be automatically sending some users' input data to Baidu's servers without their permission.
Following the uproar, Baidu Japan said it quickly fixed "the bug" and tried to reassure users by obtaining an international standard certification that sets out the requirements of an information security management system.
For the moment, the speech function is only for Android users; iOS users will get it later this month.