Can scammers really copy a relative’s voice? Yes. Voice-cloning models can produce speech that sounds like a specific person after learning from fairly short recordings, and those recordings are often public. The defence that works is simple and old-fashioned: a code word your family agrees on in advance, and a habit of hanging up and calling back.
The technology also has legitimate uses, such as audiobooks, accessibility tools and dubbing, so it is not going away. What follows explains what a clone can and cannot do, where scammers get the audio, and how to set up a family rule that elderly relatives will actually use.
How does voice cloning work, in plain terms?
A voice-cloning system learns two things from a recording: what the person sounds like (pitch, timbre, accent, the rhythm of their speech) and how to turn any text into speech with those features. Older systems needed long, clean studio recordings. Current ones can produce a recognisable imitation from much shorter samples, especially when the audio is clear and the person speaks naturally.
It shows up in scams in two ways. In the first, the scammer types a script and the system reads it aloud in the cloned voice, which works well for voice messages. In the second, a voice-conversion tool changes the scammer’s own speech into the target’s voice in near real time, which allows a live phone call. Live conversion is harder and tends to sound flatter or slightly delayed.
Where does the audio come from? Videos on TikTok, Facebook and YouTube, livestreams, voice notes forwarded in group chats, recorded online classes, even a short “wrong number” call in which the scammer keeps the victim talking. Teenagers and young adults who post a lot of talking videos are the most exposed, and their grandparents are often the ones targeted.
What a cloned voice still cannot do
What a clone does not have is knowledge. It can sound like your son, but it does not know what the family ate last Sunday, the nickname only relatives use, or the code word you agreed on. That gap is the whole basis of the defence.
Cloned voices also struggle to hold emotion over long stretches, to get the regional tones of Vietnamese exactly right and to handle spontaneous back-and-forth. Scammers cover these gaps with noise: crying, a bad line, sirens in the background, or a “police officer” who takes over the phone.
Common voice-clone scam scripts to recognise
The accident call. A child or grandchild is hurt or in hospital and money is needed for treatment right now. The voice is upset and the line is noisy, which hides imperfections.
The detained relative. A relative says they have been stopped by the police and need money for a fine or “to settle it”, then hands the phone to a fake official.
The new number. A voice message says “my phone broke, this is my new number, save it”, and only days later comes a request for money. The first message exists just to make the new number look trusted.
The boss on a voice note. A manager’s voice in a Zalo or Telegram voice message asks an accountant to make an urgent transfer to a supplier. Voice notes suit cloning well because there is no live conversation to keep up.
The bank or delivery “verification”. A friendly voice asks you to confirm details or read out a code. Here the voice may be generic rather than someone you know, but the goal is the same: an OTP or a password.
The recording trap. Short calls with simple questions designed to get you saying “yes”, your name and a few sentences. The aim may be to collect your voice for later, so it is fine to hang up on unexplained surveys.
The demand for secrecy. The caller begs you not to tell parents or a spouse. Secrecy is a structural part of the script, because one extra phone call ends the scam.
Payment to someone else. The money always has to go to a stranger’s account, a “hospital staff member” or a “lawyer”, never to the relative’s own account. That detail alone is worth stopping for.
How to set up a family code word people will remember
A code word works only if it is agreed in person, remembered under stress and never written down anywhere online. These steps keep it practical.
- Agree on it face to face. Choose it at a family meal or during a visit, not in a group chat that could be read if one account is hijacked.
- Make it odd but memorable. Two unrelated everyday words work well, for example a fruit plus a street food. Avoid pet names, birthdays and anything visible on social media.
- Decide who asks for it. The rule is simple: whoever receives an urgent call involving money asks for the word. The caller never offers it unprompted, so a scammer cannot learn it by listening in.
- Practise once. Do a light role-play: one person calls pretending to need money, the other asks for the word. Elderly relatives remember something they have done far better than a rule they have only heard.
- Pair it with a callback. If the caller cannot give the word, hang up and call the person’s saved number, or another family member. No exceptions for “my phone is broken”.
- Change it if it leaks. If the word is ever said in front of outsiders, typed into a chat or used once in a real emergency, choose a new one at the next family meal.
- Do the same at work. Small companies can agree on a verification phrase or, better still, a rule that payment changes are always confirmed through a second, known channel.
Present the rule as protection for the caller too: “If I ever call you in trouble, ask me the word. I will not be offended.”
Protecting your own voice and your accounts
Your voice is now a kind of personal data that can be misused, but treating every recording as dangerous is neither realistic nor necessary. The practical goals are to reduce long, clean, public samples and to stop scammers from using your accounts to reach your family.
If you post many talking videos, consider keeping older or personal ones visible to friends only. Be cautious with unknown calls that try to keep you talking without a clear purpose. Some people let unknown numbers go to voicemail and call back only if the message makes sense.
Lock down messaging accounts with two-factor authentication. Many voice-clone scams begin with a hijacked Zalo or Facebook account, because a voice note arriving from a familiar account is doubly convincing. Check your linked devices regularly and never share login codes with anyone.
If your voice or account has been used in a scam, warn your contacts quickly through other channels. A short message asking people to ignore any money requests protects them while you recover the account. If money has been lost, contact the bank at once and report to the police.
Finally, avoid voice-based security where you have a choice. Voice identification was designed before cloning became easy; if your bank or provider offers confirmation through its app or a passkey instead, that is generally the stronger option.
Frequently asked questions
How much audio does someone need to clone a voice?
It varies by tool and recording quality. Some systems produce a recognisable imitation from short clips, while long convincing conversations need more and cleaner audio. Assume any public talking video could be enough for a short voice message.
Can I tell a cloned voice just by listening?
Sometimes: flat emotion, odd pauses or slightly wrong regional tones can give it away. But noise and panic hide these clues, so rely on a code word and a callback rather than your ear.
What if my parents forget the code word?
The callback rule still protects them: hang up and call the saved number or another family member. A note near the phone reminding them of the rule, not the word, also helps.
Is voice cloning illegal?
The technology itself has legitimate uses; using someone’s voice to deceive people or take money is fraud. Rules on consent and personal data are still evolving, so check current regulations or ask a lawyer about a specific case.
Should I answer calls from unknown numbers?
You can, but avoid long conversations with callers who give no clear reason. Letting unknown numbers go to voicemail is a reasonable habit if you worry about your voice being recorded.
Does a code word work for businesses too?
It helps, but a process is stronger: confirm every new or changed payment instruction through a second, known channel and require two approvers for large transfers.
