Living in a multicultural society you often come across groups of people speaking their native language. If you’re like me, and can only speak English, sometimes there is a wistful thought that you wish you had the knowledge and ability to chat with them in their tongue.
A Star Trek Universal Translator would come in handy, but such things are strictly the subject of sci-fi and fantasy, right?
Not anymore!
Microsoft Research has created what seems to be a very promising real-time voice translator that so far works reasonably well for English-Chinese translations.
On October 25th in Tianjin, China, Rick Rashid, the Chief Research Officer of Microsoft gave a presentation and demonstration of the translation software to an audience of 2000 Chinese students and academics, with a positive response.
The software was able to translate Rashid’s speech in English to Chinese text, and from there synthesise the Chinese text into Chinese speech in his own voice.
According to Rashid, the initial process takes place in two steps: the English is first translated word-for-word into its Chinese counterparts, and then reordered so the sentences make sense grammatically.
Rashid explained in his presentation and blog post, that the software works by using deep neural network technology, which is patterned after the way the brain works. This technology allows a higher rate of recognition, and reduces the word error rate for speech by over 30% compared to previous methods.
In order to produce the Chinese speech in Rashid’s own voice, he had to spend one hour ‘training’ with it, beforehand.
Results are not perfect, and Rashid admits there is much work to be done, still, as (sometimes hilarious) errors do occur. However, Microsoft envisions this software as a step towards bettering humanity, by breaking down the barriers of language between people.
A lofty goal, and one certainly worth striving for!
Click here to check out the video and see the translation software in action (from 9:00 onwards).