A human voice perfectly cloned by an A.I.

We have just gotten used to the deepfake phenomenon (the artificial intelligence able to replace a face on video with another one seamlessly) but lastly, a new artificial intelligence can even clone a human voice with its characteristics.

That’s what happened to a CEO of a U.K. firm that received a phone call from his boss who demanded him a transfer of $243,000 to a supposed supplier but the CEO was cleverly scammed.

The voice on the phone, in fact, did not belong to the boss, the chief executive, but it was a fake perfectly cloned by artificial intelligence. The trick was discovered because the scammers made a second call to ask for a new transfer with the same money. This aroused the suspicion of the company management and an investigation has immediately started but attackers haven’t been found yet.

According to experts, the voice-print would have been reconstructed starting from some audio samples of the boss, then elaborated into an Artificial Intelligence system able to simulate vocal characteristics to reproduce a human voice able to speak any desired word.

As far as it is concerned, this is the first case where an A.I. has been used to scam someone. That should alert us. From now on, when we answer the phone, should we begin to doubt even a familiar voice? How can we protect ourselves? At the moment, it doesn’t seem to be an efficient tool, but we hope for the future to have something able to recognize fake voices. Doubting seems the only weapon but sometimes looks difficult and/or inopportune asking for further information to clarify the identity of the other person.

>>>  Artificial Intelligence and the frightening Roko's Basilisk


Lyrebird is a startup working in Montreal and was born in 2017 with the goal of “creating the most realistic artificial voice”. Their algorithm can get extraordinary results even with a few seconds of vocal sample. As the amount of data samples increases, the voice sounds more perfect and closer to the original. Their machine learning model can grab what makes a human voice unique (it’s what they call vocal DNA) and use it to say whatever you want.

A demonstration of the Lyrebird algorithm

Lyrebird is named for the bird which can imitate other birds’ calls but also other kinds of sounds.

The whole thing raises ethical issues. How does this technology affect our lives? Will we doubt every vocal message or phone call? It’s possible. Anyway, this A.I. along with deepfake videos will be able to generate totally fake videos with people doing and saying things they never did or said.

However, it will be a great opportunity for cinema and music to bring to life actors and singers who are now gone.

A demonstration of how the lyrebird imitates sounds

Try Lyrebird’s app and see its potential!

Lyrebird now is called Descript.