Google working on voice-activated universal translator

Douglas Adams' Babel fish may not exist outside of the fictional universe, but Google is working on the next best thing: a real-time universal language translator.

Google is looking to move its translation tool into the real-time, offline world with an app that translates your speech in real time, German magazine Der Spiegel has reported.

The software is under development under the leadership of Google Research's Franz Josef Och. Although the Google Translate app works in real time to translate text to text and voice to text, Och wants to build the perfect translation device based on artificial intelligence.

Such a device could listen to its user speak a word or phrase, then repeat it back translated into the chosen language; or, speaking into a phone, the words spoken into one handset could be translated in transit so that the listener on the other end hears their own language.

Yet it's not all smooth sailing. Languages around the world can be structured very differently. Interestingly, neither Och nor any members of his team are linguists; instead, they apply maths and statistics to the problem of translation, building algorithms that can correlate existing translations and find the most accurate. To this end, Google Translate allows users to rate translations so that Google can collate data on accuracy.

However, Google is not the first, and may not even be the most advanced, in creating such a tool. Microsoft already demonstrated speech recognition and translation software in November last year. Chief research officer Rick Rashid gave a demonstration in Tianjin, China, of software that translated his speech from English to Chinese, playing the translation back in Rashid's own voice — collated from voice samples. Skip to around the six-minute mark in the below video for the demonstration.

Google, however, has another advantage above and beyond what it has already achieved with Translate: Google Now. The voice search software can learn and adapt to the user's syntax for more accurate results; the more the user communicates with Google Now, the better the software becomes at understanding his or her speech patterns. Its conversational approach to voice commands — for example, following "What is the time in New York?" with "Is it raining there?" — also gives it a potential edge.

Although Google Translate isn't perfect, with hundreds of millions of people using it per year, it can only get better — and Google's suite of existing and upcoming products could play a part. Imagine Google Glass translating the language around you and translating it directly back to the earpiece. It's a few years away yet, but whoever gets there first, software of this kind will absolutely revolutionise global communications.