AI voice replication – now this is creepy

‘Basically, all you have to do in order to replicate someone’s voice is feed the program an audio clip as short as two seconds long. Voicebox will then “match the audio style,” and boom: with little more than a written prompt and the click of a few buttons, you can get an AI-powered replica of your friend or family member’s voice — and the ethical and legal implications are palpable.’

https://futurism.com/the-byte/facebook-ai-replicate-friends-voices

That quote comes from a Futurism article looking at the new AI announced by Meta [Facebook]. The AI, called Voicebox, is not being made available just yet, because even Meta is aware of how much abuse it could cause. There are already documented instances of parents receiving messages from their kids in which the kids say they’re in trouble and need urgent funds. The parents send money only to discover that they have been scammed. And those are text messages.

Now, imagine if your son or daughter rang you and begged for money to avoid some terrible fate… Unless that offspring were standing right next to you in real time, how would you know if their voice was real or faked? I feel ill just thinking about it.

I’ve imagined the future with AI quite a bit, but somehow I always assumed that no AI would be capable of faking someone’s voice. Or at least, not well.

Of course, we have only Meta’s word on how well Voicebox can fake a given human’s voice. I suspect that a two second exposure to a human’s voice would not give the AI enough material to work with when it comes to whole sentences. Our voices change, depending on how we feel. For example, we can generally hear when someone is smiling, or angry, or in some other, more subtle mood.

Yes, an AI could be told to fake ‘fear’, but how would it know how that particular person sounds when they’re afraid? Generic fear, the type you see in movies, is rarely what real fear sounds like. Unfortunately, a parent might not know what their son or daughter sounded like when deathly afraid either.

At the moment, the best way to avoid being scammed is to go directly to the source, but what if that source is thousands of kilometres away? What if they don’t answer their phone?

I fear that when this particular genie slithers from its bottle, we’ll be forced to give each other ‘passwords’ to prove that the voice on the other end of the phone really is who we think it is. What an appalling prospect.

Meeks