The phone rings. On the other end is your son's voice, panicked: he has had an accident, he needs money right away, he cannot explain. The voice is his, the tone is his, even the way he addresses you. And yet it is not him. Speech synthesis technology has become accessible enough for a fraudster to imitate a voice starting from short recordings published online. It is one of the most insidious forms of AI manipulation, because it targets a sense we instinctively trust.
How voice cloning works
Speech synthesis systems analyse recordings of a person and learn to reproduce their timbre, rhythm and pronunciation. With plenty of material the result is more faithful; but even with little, a synthetic voice can be convincing in a short, noisy, emotionally charged phone call.
Where does the audio come from? The sources are often public:
- videos and stories on social media in which the person speaks;
- voice messages forwarded in groups;
- interviews, podcasts, recorded presentations;
- a personalised voicemail greeting.
So the fraudster does not need access to your phone: what is already visible to everyone is enough.
The most common schemes
The family emergency
A voice that sounds like a relative asks for money for an emergency: an accident, an arrest, a problem abroad. Often a second person then takes over, posing as a lawyer, doctor or police officer and giving payment instructions.
The executive asking for a transfer
In a business setting, a voice imitating a manager calls someone in accounts and requests an urgent, confidential payment to a supplier. The demand for secrecy is there to prevent checks.
Voice confirmation
Some fraudsters want the opposite: to record your voice saying words like "yes" or "I confirm", to reuse in other contexts. That is why it is wise not to answer suspicious calls asking for generic confirmations with single words.
The warning signs
The voice may be perfect, but the scam pattern almost always gives away a few elements:
- extreme urgency: you must act now, there is no time to call anyone;
- a demand for secrecy: "don't tell Dad", "don't mention it to your colleagues";
- unusual payment methods: prepaid cards, cryptocurrency, transfers to never-used accounts, couriers coming to collect cash;
- a call from an unknown number with an excuse to justify it ("I lost my phone");
- a short, confused conversation, with interruptions or noise that make it hard to ask questions.
The defences that really work
Trying to recognise a synthetic voice by ear is becoming less and less reliable. Effective defences are organisational.
Call back yourself
Hang up and call the person back on the number you already know, not the one they called from. If they do not answer, call someone close to them. No real emergency is jeopardised by two minutes of checking.
Agree on a safe word
In your family, choose a word or question that only you know and that appears nowhere online. In case of an urgent request, the caller must know it. It is a simple and very effective solution, especially for protecting elderly people.
Ask personal questions
Ask something only the real person would know and that cannot be found on social media: the name of their first pet, what you last ate together. A fraudster will tend to dodge the question or push the urgency.
At work, double confirmation
No urgent payment requested by phone should be made without confirmation through a second, independent channel, such as an internal email or a call to the known company number. This rule must apply also, and above all, when the request seems to come from senior management.
Reducing your voice's exposure
It is not realistic to stop appearing in videos, but you can limit the risks: make the profiles where you post personal content private, avoid leaving long voice messages in large groups and use a standard voicemail greeting rather than one recorded in your own voice.
If you have been scammed
Contact your bank immediately to try to block the transaction, keep the number that called you, the time and any related messages, and file a police complaint. Also warn the person whose voice was imitated: there may be further attempts targeting their contacts.
In short
- A few seconds of public audio can be enough to imitate a voice convincingly.
- Urgency, secrecy and unusual payment methods are the most reliable signs of a scam.
- Always call the person back on the number you already know before acting.
- A safe word agreed within the family is a simple and effective defence.
- At work, no urgent payment without confirmation on a second channel.