From the guy rope who does the voice - over for movie trailers to the announcers on the subway , our lives are full of faceless voice . And while most of us are content to ramp up a mental image of these discorporate orators , a group of researchers from MIT has cash in one’s chips a tone further by creating anartificial intelligencesystem that can rebuild people ’s face just by listening to their vocalization .
The app , called Speech2Face , is a cryptical neural internet that was trained to acknowledge the correlativity between voice and facial features by keep millions of YouTube video of people talking . In doing so , it learned to associate unlike aspects of the audio waveform with a speaker unit ’s long time , grammatical gender , and ethnicity , as well as certain cranial feature such as the shape of the head and the breadth of the nozzle .
When the research worker then fed the system audio recordings of people ’s voices , it was capable to engender an image of each speaker’sfacewith reasonable truth .
Obviously , characteristic like hairstyle , facial haircloth , and sure other elements of forcible visual aspect are impossible to predict from a mortal ’s voice , so the developer insist that their finish was “ not to predict a recognisable ikon of the exact face , but rather to capture dominant facial traits of the person that are correlated with the inputspeech . ”
In a paper published onIEEE Xplore , the researcher say this technology could one Clarence Shepard Day Jr. discover a range of useful applications , such as generating faces for picture calls without the penury for cameras .
However , some improvements are clearly still needed , as while the images make by Speech2Face are generally a good couple for nerve type , they often only bear a general resemblance to the utterer . The system is also prostrate to the occasional mistake , with roughly 6 pct of the faces it make being of the wrong gender , and some of the awry ethnicity .
Nevertheless , faceless voices are one step near to becoming a thing of the past , which should have major implication for prank caller-up at least .
