How to Tell if a Voice Is AI

How to tell if a voice is AI: can you hear the difference?

Sometimes, but you should not count on it. Knowing how to tell if a voice is AI starts with accepting that a well-made clone can fool a parent, a colleague or a bank employee, especially over a phone line.

So split the problem in two. In the moment, verify the person through another channel. Afterwards, if you have a recording, examine it with your ears and with an AI voice detector.

How does voice cloning work?

Voice cloning works by teaching a speech model the characteristics of one person: pitch, accent, rhythm, the way they stretch certain sounds. The model then reads any new text in that voice.

The sample can come from anywhere: a video posted online, a voicemail greeting, a podcast, a recorded call. The US Federal Trade Commission warns that scammers use voice cloning to make requests for money or information more believable, notably in fake family-emergency calls.

Some tools turn typed text into speech; others convert a live speaker into the target voice in near real time. That is why a cloned caller can sometimes answer questions instead of playing a fixed message.

Caller ID does not help either. Phone numbers can be spoofed so that a call appears to come from a known contact, and a voice message can arrive from a hacked account that really belongs to your friend. Treat the voice, the number and the account as claims to verify, not as proof of identity.

What audio tells suggest deepfake audio?

The audio tells of a cloned voice are about what is missing: natural breathing, room sound and emotional variation. They are worth knowing, but each one has an innocent explanation.

Bottom line: the tells raise or lower your suspicion. They do not settle anything, and their absence does not mean the voice is real. If you have a recording, listen to it twice with headphones, once for the voice and once for everything around it: room sound, background noise and the moments between words.

What does an AI voice scam sound like?

An AI voice scam sounds like a familiar person in an unusual hurry. The voice is new technology, but the script is old: it is built to stop you from thinking and checking.

The same scripts target companies, where a cloned executive asks finance to pay a supplier urgently. Our page on voice clone scams walks through these scenarios, and fraud and KYC teams will find controls for onboarding and payment approval.

Where do cloned voices show up?

Cloned voices show up wherever a familiar voice can unlock trust: family calls, workplace requests, customer service and voice messages. Knowing the common settings helps you decide in advance what you will always verify.

The pattern across every row is the same: the verification channel must be one you chose, not one the caller gave you. A scammer who controls the phone number, the link and the “colleague” you are told to contact controls the whole conversation.

The call-back rule and the family code word

The call-back rule is the best defense against a cloned voice: hang up and call the person back on a number you already have, never on the number that called you. It works whatever the quality of the clone.

1. 01Say you will call right back, and hang up. A genuine relative or colleague will understand.
2. 02Call the person on the number saved in your phone or listed in your own records.
3. 03If you cannot reach them, call someone who would know where they are.
4. 04For a company request, confirm through a second channel and a second person before any payment.

A family code word adds a second layer. Agree on a word or question with close family, in person, and never post it online. Anyone calling in an emergency must give it. Businesses can do the same with a verification phrase for payment requests.

How do you know if a voice message is AI with a detector?

You check a voice message with a detector by uploading the original recording and reading which segments it flags. Detectors are most useful afterwards: on a voicemail, a voice note, or a call you were able to record.

The AnyFact AI voice detector accepts MP3, WAV, M4A, OGG, WebM and OPUS files up to 25 MB and 5 minutes. An open-source anti-spoofing model based on wav2vec2 scores each segment, Whisper transcribes the speech, and the transcript is checked for scam patterns: urgency, secrecy, money requests and changed bank details.

Read the verdict as an indication. Re-recorded audio, speaker-phone captures and heavy compression all weaken the signal, which is why the result always lists its limits. Our guide on whether AI detectors are accurate explains what affects reliability.

• Export the original file from the messaging app rather than recording it with another phone.
• Keep the sender’s number or account name and the time it arrived, alongside the file.
• If the message is long, upload the part where the request is made: that is where the scam patterns appear.
• Do not forward the message to other people “to check” before you have verified it.

What should you do if you are targeted by an AI voice scam?

If you are targeted, stop the conversation, verify independently and act quickly on any payment. Embarrassment helps scammers; reporting helps everyone.

1. 01Do not send money, codes or personal details, and do not call back the number that called you.
2. 02Contact the real person through a channel you trust.
3. 03If you already paid, call your bank or payment provider straight away and ask them to stop or recall the payment.
4. 04Save the evidence: the number, the time, any voice message and any messages that followed.
5. 05Report it to the police and to your national consumer protection or fraud reporting body. In the US, that is the FTC.
6. 06Warn family members or colleagues who could receive the same call.

If your bank or another service lets you sign in by voice, ask whether you can add or switch to another verification method. A voice that can be cloned should not be the only key to an account.

Video calls can be faked too, with a cloned face and voice together. See how to spot a deepfake video for the visual checks, or run a suspicious clip through the deepfake detector.

Checklist: is this voice real?

Keep this checklist somewhere you will see it, and share it with older relatives and with anyone who approves payments at work.

1. 01Is the caller asking for money, codes, passwords or new bank details?
2. 02Am I being rushed, or asked to keep it secret?
3. 03Did I verify the person on a number or channel I chose myself?
4. 04Did they give the family code word, or answer a question only they would know?
5. 05Does the voice sound flat, too clean, or oddly timed for the situation?
6. 06If I have a recording, did I save it and check it with an AI voice detector?

If the answer to either of the first two questions is yes, and the third is no, stop. That combination is the signature of a scam, whatever the voice sounds like.

Frequently asked questions