Cloned voice and video

Threats

A voice is convincingly cloned from fifteen seconds of audio. What counts as proof now, and what to agree with your family in advance.

“Call and check the voice” worked for decades. It has stopped being reliable: a voice is cloned from a short recording, and the recording comes from your own posts.

15seconds
Of audio is enough to synthesise a recognisable voice

The material comes from stories, voice messages, a video from a party, a talk at work. Nobody needs to record you separately.

How it is used

  • A call “from a relative” in trouble. The voice is familiar, the line is bad, the request urgent — bad audio conveniently masks the flaws in the synthesis.
  • A voice message from a hijacked account. More convincing than text, and therefore more effective.
  • A call “from your manager” asking you to pay an invoice urgently. A scheme aimed at companies, but it reaches small businesses too.
  • A video call with a swapped face. Still rare and requiring preparation, but no longer science fiction.

Signs of a fake

What you noticeWhy it happens
Even intonation with no pauses for thoughtSynthesis reads a text; it does not converse
No answer to an unexpected questionThe script was written in advance
Insisting the line is badNoise hides the flaws in the synthesis
Avoiding a video callVideo is noticeably harder to fake
Odd stress in names and place namesWords outside the training data
Never laughs, never interruptsLive speech does not sound like that
These signs help, but do not rely on your ear alone: synthesis keeps improving and lists of tells go stale.

What to do instead of recognising a voice

  1. Hang up and call back yourself

    To the number in your contacts, not the one that called. This is still the most reliable check, and synthesis does not get around it.

  2. Ask something with no answer online

    Not “what is the dog called” — that is on social media. Something like “what did we eat last Sunday” or “where is the blue mug”.

  3. Agree a code word

    One word for the family, spoken with any request for money. Choose it in advance and do not store it in a chat.

  4. Take a pause

    Every scheme breaks on delay. “I will call you back in ten minutes” is a sufficient response to anything.

What lowers the risk in advance

Provides material
  • Public stories with your voice
  • Voice messages in public chats
  • Recordings of talks in the open
  • A public list of relatives in your profile
Makes it harder
  • A private profile and a limited audience
  • A family code word
  • The habit of calling back yourself
  • Having this conversation with older relatives

The schemes that use a cloned voice are covered in messenger scams, and the conversation with parents separately.

Copied