Microsoft's speech translation tech gets 30 percent accuracy improvement

Microsoft's Chief Research Officer, Rick Rashid, recently gave a demonstration of a technology it's jointly developing in China. First introduced in March, the software will take speech in a number of 26 languages, including Mandarin Chinese, and translate it to spoken English using the user's own voice. The October 25th demonstration revealed Microsoft and its partners have refined the technology to be more accurate, with an error rate reduced by about 30 percent, with one word in 7 or 8 being correct.

"Of course, there are still likely to be errors in both the English text and the translation into Chinese, and the results can sometimes be humorous," Rashid wrote. "Still, the technology has developed to be quite useful."

The tech relies on research into Deep Neural Networks technique that uses human brain behavior patterns performed along with the University of Toronto. The software first needs to learn a user's voice by recording about an hour worth of phrases so it can then recreate his or her voice as closely as possible. The system was also programmed with a few hours of speech from a native Chinese speaker.

A video of Rashid's October 25th presentation at the Microsoft Research Asia's 21st Century Computing event is embedded below. He used the system to speak in English, with the software presenting it in Chinese. [via TechCrunch]

By Electronista Staff

Post tools:

TAGS :  

industry, software, Microsoft, University of Toronto