Imagine you’re sitting quietly at home when suddenly your phone rings. You pick it up and hear the voice of your grandson or granddaughter on the other end. Their voice is trembling, perhaps a little hoarse with emotion, but you have no doubt whatsoever – you recognise it straight away. You hear a plea for urgent financial help because there’s been an accident. Fear sets in straight away, along with a natural urge to save a loved one, because after all, you can hear them with your own ears and they’re undoubtedly in need, aren’t they?
Unfortunately, this could be a very clever trap. These days, fraudsters no longer need to impersonate someone else or alter their tone of voice.
Deepfake audio, often referred to as voice cloning, is a modern method of almost perfectly replicating the way a specific person sounds. It is a form of fraud in which criminals use computer programmes to create a recording that sounds like the voice of someone we trust – for example, a family member, a friend or a boss.
Fraudsters need only a short audio clip (e.g. from a video posted on social media) to ‘train’ a computer to mimic a person’s intonation and tone of voice. This enables them to ring us and ask for urgent financial help or to reveal passwords, whilst sounding so convincing that it can be really difficult to become suspicious.
The whole process usually goes like this:
- Collecting samples: The fraudster first needs to obtain a voice recording of the person they wish to impersonate. These days, this is easy – all it takes is a short video we’ve uploaded to the internet (e.g. a holiday clip or a birthday message).
- Training: A special computer programme ‘listens’ to this recording thousands of times per second. It learns every nuance in the voice – how the person breathes, how they emphasise words and the timbre of their voice. Based on this, the computer creates a ‘map’ of the sound of that specific voice.
- Real-time transformation: When a fraudster calls you, they speak into their microphone in their own voice. However, software on their computer instantly – in a fraction of a second – alters their words so that they sound like the voice of the person they are trying to impersonate.
It’s a bit like a criminal putting on a special costume that transforms everything they say into the voice of someone close to us. They say whatever they like, whilst we hear a familiar voice on the other end of the line – one we trust.
It’s worth knowing that deepfakes aren’t necessarily just pre-recorded videos that someone sends us in a message or posts online. Today’s technology allows fraudsters to superimpose a ‘digital mask’ onto a face in real time – that is, during a video call that’s currently taking place. This means you could be chatting to someone via a messaging app (e.g. WhatsApp), see your friend’s face and hear them answering your questions, whilst in reality a criminal is sitting on the other end, and their voice and image are being altered in real time by a computer.
What’s more, fraudsters can modify not only their own face, but also what is visible behind them. A fake background can give the impression that someone is calling you from a reputable institution, such as a bank branch, a police station or a hospital ward. All this is done to gain your trust and make the situation appear official and genuine.