Alibaba’s Smart Speaker is Learning Dialects

There are many Chinese dialects Alibaba's voice assistant can't recognize -- they're hoping to solve that with a recent investment of $15 million

1 0
12:34 AM HKT, Sat March 23, 2019 1 mins read

Vast databases undergird the system of Tmall Genie, Alibaba’s intelligent voice assistant and answer to Amazon’s Echo, providing it with encyclopedic information and billions of real dialogue sessions. But there’s a gap in the smart speaker’s knowledge that Alibaba’s AI Labs is hoping to patch up with a recent R&D investment of 15 million USD: Chinese dialect recognition.

According to Alibaba’s official press blurb, the Genie is “the top-selling voice-controlled speaker in China, the only market where it is currently available. Powered by AliGenie, the device can understand commands in Mandarin Chinese, the most widely used language in the world.”

Except that to huge swaths of the country, Mandarin isn’t necessarily their go-to language. Although Chinese language is usually categorized into five broad groups, there are as many as 200 individual dialects in use in the country. And they’re rarely mutually intelligible — someone speaking in the Minnan dialect common in parts of Fujian province is unlikely to be understood by a Cantonese speaker, for example.

To tackle this project, Alibaba is partnering with linguistic experts, universities and other institutions, and has already begun compiling a dialect database based on Tmall Genie dialogue samples. They’ll first focus on the Sichuan dialect and then branch out from there.

Hey, maybe Kafe Hu can help?

Alibaba claimed in January to have sold 5 million Tmall Genies since 2017, and Xiaomi’s voice assistant XiaoAi — which Xiaomi is also reported to be working on dialect recognition for — has been pushing the bounds with new features like a touchscreen.

Competition is clearly hotting up, and though it seems a little odd to announce a major investment into helping your technology learn more languages in the same week that your flagship delivery service was accused of using its app to secretly listen to users’ conversations, the dialect recognition feature would be a valuable addition to a product that only seems set to grow in popularity in China — especially as the Genie program is being inserted into BMW cars.

Cover photo: Alizila

Join the Conversation
Write comment

Pour yourself a stiff one, we'll be with you in a minute