Voice cloning means using AI to copy a real person’s voice, so a computer can say new things that sound like them. It has honest uses, such as dubbing a film or giving someone a familiar voice to read their messages. It is also used by scammers to fake calls from family members.
Think of a gifted impressionist who only needs to hear you speak for a moment before they can do your voice, saying anything they like. Voice cloning is that impressionist, as software. It can copy how you sound; it cannot copy what you would actually say.
In a sentence: “The film studio used voice cloning, with the actor’s permission, to dub his lines into Spanish in his own voice.”
What does voice cloning mean?
The plain point: an AI learns what makes one voice sound like that person, then speaks new words in it.
It builds on text-to-speech, the technology that reads written words aloud. Ordinary text-to-speech uses a stock voice. Voice cloning makes a voice that matches a particular person from a recording of them. You type the words; the clone says them.
How little audio does it take?
The plain point: sometimes just a few seconds.
In 2023, Microsoft researchers described a model called VALL-E that could copy an unfamiliar speaker’s voice from a three-second recording.1 The US Federal Trade Commission (FTC) has warned that a scammer needs only “a short audio clip of your family member’s voice”, which could come from videos posted online.2
A familiar voice on the phone is no longer proof of who is calling. A short public clip can be enough to copy it.
Who is allowed to clone a voice?
The plain point: responsible services ask for the speaker’s permission first; scammers do not.
Microsoft, which sells custom AI voices to businesses, keeps the feature behind an application process and says its customers “must get sufficient consent” from the person whose voice is used.3 In the US there is no federal law on copying someone’s voice yet: the NO FAKES Act, which would add one, was reintroduced in May 2026 and was still a bill, not law, when this site last checked on 12 Nov 2026.4 Until then, protection comes from state laws such as Tennessee’s ELVIS Act, which reaches conduct connected to Tennessee rather than the whole country.Not read at source: Tennessee’s ELVIS Act (signed 21 Mar 2024, in force 1 Jul 2024), as recorded on this site’s voice and likeness law page from coverage by ASCAP, NPR and several law firms read 11 Sep 2026; the state’s own page at tn.gov did not load. That page notes it is state law that binds conduct connected to Tennessee. Cloning a voice or a face: what the law says explains it.
How voice-cloning scams work
The plain point: the scam relies on panic, so the defence is to slow down and check.
The usual pattern is a call that sounds like a relative in trouble, asking for money fast. The FTC’s advice: “Don’t trust the voice. Call the person who supposedly contacted you and verify the story. Use a phone number you know is theirs.”2
In the US, the Federal Communications Commission ruled in February 2024 that robocalls using AI-generated voices count as “artificial” voices under its robocall law. That brings them under the same rules as other robocalls, including the need for telemarketers to get people’s written consent first.5 A law does not stop a criminal from calling, though, so the habits matter more:
- Agree a family code word and ask for it in any emergency call.
- Hang up and call back on a number you already have.
- Treat urgency and secrecy as the warning sign, especially with requests for gift cards, crypto or bank transfers.
Deepfake self-defence walks through the whole routine.
Often confused with
- Text-to-speech: the general job of reading text aloud, in any voice. Cloning is the version that copies one real person. See text-to-speech meaning.
- Deepfake: a fake made to fool people. A cloned voice becomes a deepfake when it is used to pass off words as really spoken by that person. See deepfake meaning.
- Speech-to-text: the opposite job: listening and writing down. See speech-to-text meaning.
Questions people ask
How much audio do you need to clone a voice?
Not much. Microsoft’s VALL-E research used a three-second recording,1 and the FTC warns that a short clip posted online can be enough for a scammer.2
Is voice cloning legal?
It depends on permission and on where you are. Legitimate services work with consent: Microsoft says its customers “must get sufficient consent” from the person whose voice is used.3 In the US, robocalls made with AI-generated voices count as “artificial” voices under the robocall law.5 For the rest, this site’s voice and likeness law page explains how the rules vary from place to place.
Learn more on this site
- The scams: how AI fraud actually works.
- Labels on AI content: watermarking meaning.
- Staying safe: staying safe with AI.
- The words: the AI glossary.
- Related: How to Make a Talking Photo or Lip-Sync Video With AI, With Consent.
- Related: How to Write an ElevenLabs Voice Design Prompt.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is voice cloning?
- Writing down what someone says
- Using AI to copy a real person’s voice
- Reading text aloud in a stock voice
Show the answer
Answer: Using AI to copy a real person’s voice. The page says voice cloning makes a voice that matches one particular person, so a computer can say new things that sound like them.
2. How much audio can it take to copy a voice?
- Sometimes just a few seconds
- Many hours of studio recordings
- A full day of the person talking
Show the answer
Answer: Sometimes just a few seconds. The page says Microsoft researchers copied a voice from a three-second recording, and the FTC warns a short online clip can be enough.
3. A call sounds like a relative in trouble, asking for money fast. What should you do?
- Trust it, because the voice sounds just right
- Do what they ask quickly, before it gets worse
- Hang up and call back on a number you already have
Show the answer
Answer: Hang up and call back on a number you already have. The page says a familiar voice is no longer proof of who is calling, so slow down and check on a number you know.
Related terms: AI ethics · AI governance · Red-teaming · Shadow AI · Slop · Watermarking
- 1Wang et al. (Microsoft), Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers, arXiv:2301.02111, submitted 5 Jan 2023, read at source 5 Oct 2026“Vall-E can be used to synthesize high-quality personalized speech with only a 3-second enrolled recording of an unseen speaker as an acoustic prompt.” ↑a“Vall-E can be used to synthesize high-quality personalized speech with only a 3-second enrolled recording of an unseen speaker as an acoustic prompt.” ↑b
- 2US Federal Trade Commission, Scammers use AI to enhance their family emergency schemes, 20 Mar 2023, read at source 5 Oct 2026“All he needs is a short audio clip of your family member’s voice — which he could get from content posted online” ↑a“Don’t trust the voice. Call the person who supposedly contacted you and verify the story. Use a phone number you know is theirs.” ↑b“All he needs is a short audio clip of your family member’s voice — which he could get from content posted online” ↑c
- 3Microsoft, Text to speech transparency note, 31 Mar 2026, read at source 5 Oct 2026“Customers must get sufficient consent under all relevant laws and regulations from the voice talent to use their voice to create a custom neural voice.” The page also describes the service as “a Limited Access service, and registration is required for access to this service.” ↑a“Customers must get sufficient consent under all relevant laws and regulations from the voice talent to use their voice to create a custom neural voice.” The page also describes the service as “a Limited Access service, and registration is required for access to this service.” ↑b
- 4Congress.gov, S.4591 and H.R.8915 (the NO FAKES Act), both read at source 11 Sep 2026 and re-checked in the GovInfo bill status records 12 Nov 2026, as cited on this site’s voice and likeness law page: reintroduced 20 May 2026; S.4591 ordered reported out of committee on 18 June 2026; H.R.8915 still at the first stage; pending, not enacted.↑
- 5Federal Communications Commission, FCC Makes AI-Generated Voices in Robocalls Illegal, 8 Feb 2024, read at source 5 Oct 2026the FCC “announced the unanimous adoption of a Declaratory Ruling that recognizes calls made with AI-generated voices are ‘artificial’ under the Telephone Consumer Protection Act (TCPA).” … “Under FCC rules, it also requires telemarketers to obtain prior express written consent from consumers before robocalling them.” ↑athe FCC “announced the unanimous adoption of a Declaratory Ruling that recognizes calls made with AI-generated voices are ‘artificial’ under the Telephone Consumer Protection Act (TCPA).” … “Under FCC rules, it also requires telemarketers to obtain prior express written consent from consumers before robocalling them.” ↑b
Wang et al. (Microsoft), Neural Codec Language Models are Zero-Shot Text to Speech Synthesizers, arXiv:2301.02111, submitted 5 Jan 2023, read at source 5 Oct 2026.
The comparison on this page (an impressionist), the honest-use examples and the safety habits (which repeat the advice on this site’s deepfake pages) are this site’s own ways of explaining, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK