Yet more technology (like these real tractor beams) that seems inspired by the wonderful Star Trek Universe. It seems Microsoft have started developing a Universal Translator-esque system that outputs translated text in real time…in the voice of the speaker.
Obviously, speech-to-text recognition software isn’t exactly brand new – Siri has become quite an established example (though she is, of course, one of many and not even the first of her kind). However, the issue (and what many a meme has made fun of) is that they often get it wrong – about every four to five words according to Rick Rashid, Microsoft’s research chief.
So how is Microsoft’s Bing Translate system better? It uses something called “novel neural networking” – that is, the machine can learn to recognize errors and reduces them to one in seven or eight. Not exactly perfect, but definitely on its way to significant improvement.
Probably the coolest thing is the voice though – it seems that Rashid spent a full hour training the novel neural networking algorithm so that the voice could retain the original speaker’s vocal cadence. This is important in the translation of languages, because the way something is said can have a profound effect on the meaning.
Skip to 7:35 on this video to check out the translation in action.
So it isn’t quite perfectly the same yet – but it’s definitely on its way. It seems we could easily have bilingual presentations without the need of a translator very soon.
Now we just need to work on a way to get it to detect the language someone is speaking, override their own speech and instantaneously allow us to hear what they’re saying in our own language. Then shove it onto a badge and we’re good to go!