Android to translate spoken words in real-time in the near future

Gone are the days when translation systems are a source of laughter-induced stomach aches. While there are still some rough corners, advancements in artificial intelligence and machine learning have significantly improved translations to an almost eerily terrifying degree. Google is unsurprisingly at the head of the pack and it has just demonstrated how Google Translate on Android might, in the not so distant future, translate what you say on the fly.

Google Translate is already quite a handy tool that can do more than just translate the text you paste into it. Thanks to computer vision and, of course, machine learning, it can not only translate the text your phone's camera is "seeing" but also overlay the translation on top of the original. What it's missing, though, is perhaps the holy grail of translation, transcribing and translating spoken words in real-time.

That's exactly what Google demonstrated to a handful of folks in San Francisco, showing off its chops at AI and natural language processing. The simple process, which involved speaking through the Android phone's mic and hearing the translated output immediately, belies the sophisticated transcription, processing, editing, and embellishing that the translator does behind the scenes.

That complexity is also why this real-time translation feature will require an active and probably speedy Internet connection. That goes contrary to Google's recent push to confine translation on the device, both for privacy and performance reasons. At this point, it considers real-time multilingual transcription to be beyond the capabilities of most smartphones.

At the same time, it only works on voice audio that is recorded live rather than being played on the phone from a media file. The latter requires a different kind of neural network models that Google isn't ready to tackle yet. Other than a very general "near future", the company didn't make any promises when the future would roll out even to testers.