Don't put words in my mouth that I didn't say is a phrase repeated in thousands of arguments, but what if it were possible to do exactly that? Let's think about it for a moment: Today Photoshop shows us things that don't exist. We've seen initiatives with the potential to digitalize actors and bring them back to the big screen. Assistants on mobile devices recognize our voice perfectly. How hard would it be to imitate it then? Adobe's Project VoCo is trying to find out.
The MAX conference that Adobe organizes every year allows developers and consumers alike to get a closer look at some of the things the company is working on, whether they are products ready for the market or relatively obscure prototypes. Project VoCo fits into this second category, a name that doesn't tell us much, but which carries a chilling potential. What kind of potential is that? Well, we all know what Photoshop can do in the hands of an expert. Models don't have a single imperfection on their skin, cars that haven't been built yet perform spectacular stunts on the beach, and politicians lose ten or fifteen years when they're campaigning. Now, think about those resources applied to audio. Take someone's speech, edit it, replace and place words that aren't there.
What is Project VoCo?
That's what Project VoCo is about. In the words of Adobe developer Zeyu Jin, VoCo does for audio what Photoshop does for photography. The software is something like a pre-alpha at the moment, but if we follow the brief demonstrations, all the user has to do is enter text into a small dialog box, and VoCo will play it using exactly the same voice. Of course, such a function implies a very broad analysis of the audio, and several sources indicate that VoCo needs at least 20 minutes of content to produce a faithful reproduction. If we calculate how long politicians' speeches last on average, I'm convinced that VoCo has more than enough material.
Ethical concerns
While Adobe has legitimate reasons to move forward on a project with these characteristics, ethical alarms have not stopped sounding. As always, the blame is not on the tool, but on the one who uses it. If VoCo falls into the wrong hands, a logical and coherent speech could be turned into aggressive or belligerent... and vice versa. It's already very difficult for us to detect edited images and fake videos. Imagine extending that same conflict to an audio piece that can last a couple of hours. Even so, we will keep an eye on VoCo.
