Baidu Inc. today announced the availability in beta test of a new “speech-to-speech simultaneous translation service” designed to facilitate communication at business events.

The Chinese technology giant said the service can provide highly accurate, real-time translation of speaker presentations at events in both audio and text.

The way it works is interesting, too, as the translations are delivered directly to people’s personal devices. Users simply download the Baidu app, scan a QR code for the specific conference they’re attending, and the translations will be streamed directly to their device.

The simultaneous translation system is built on Baidu’s newly-developed context-aware translation model, a method that enables the system to generate smooth and coherent translations by constantly referencing to previously generated translations. It introduces a novel concept called information units to intelligently divide speech into semantic unit in real time, in turn using those divisions as functional units for translation. This unique strategy was inspired by human interpreters who segment speech input into meaningful units, so as to better ensure quality and reaction time.

Hua Wu, chief scientist of Baidu’s natural language processing unit, said the company isn’t trying to replace human interpreters but rather to make translation available to those who don’t have access to their services. He noted that a machine that can translate in real time and read speech out loud was “once regarded as science fiction.”

Constellation Research Inc. analyst Holger Mueller praised Baidu’s progress in simultaneous translation, since language barriers remain a big burden to human communication, understanding and access to culture. He said he was particularly interested in the commercial applications of the new service.

“All major cloud and AI vendors are offering translation services in one form or another, but mostly to consumers,” Mueller said. “It’s the business-level grade that is the next milestone.”

Photo: Cloud Computing/Flickr