Yes Star Trek fans, you may pause to swoon, but it's also likely even non-Trekkies will find the latest development in the quest for a viable universal translator to be interesting.
(http://mimg.ugo.com/201005/45046/cuts/universal-translator_288x288.jpg)
The system works by recognizing a person's words, and then converting the speech into organized sentences — in this case, Chinese. This data is then picked up by speech synthesizing software trained to replicate the speaker's voice and their unique cadence.
It's an upgrade from Microsoft's earlier technology that modified synthesized speech to match a person's voice, but could only speak typed text. The new software is modeled on how networks of brain cells operate, and takes an hour or so to train itself to process a particular person's speech patterns.
In a Microsoft blog post about the new system, Rashid says exploring the new technology mimicking neural networks is responsible for the significant jump in the company's software capabilities, with the error rate dropping from one word in four or five being incorrect to just one in seven or eight.
Fortunately, this is a case where we can see it to believe it. Microsoft's chief research officer demoed the technology to an audience in China late last month, as the video below shows.
www.youtube.com/watch?feature=player_embedded&v=vGYKxrRWDKo
the cool translator part happens just after the 6:00 mark. but the whole thing's not too bad.