How to tell if a voice note is AI-generated

A voice note arrives on WhatsApp. It sounds like your brother, your boss or your bank manager, and it is asking for something urgent. Before you act, it is fair to ask: is this voice message AI?

This guide explains what you can and cannot hear, which clues matter more than the sound itself, and how to check a voice note sensibly. The short version: listening helps a little, a detection tool helps a bit more, and verifying the person through a separate channel is what actually keeps you safe.

Why this question matters now

Voice cloning software can copy how someone sounds from a short recording. The US Federal Trade Commission warned in its consumer alert on AI-enhanced family emergency schemes that a scammer may need only a short audio clip of a family member, which could come from content posted online, plus a voice-cloning program. Its key advice was blunt: don't trust the voice.

That advice is the core of this guide. A voice is no longer proof of identity on its own, however familiar it sounds.

What you might notice when listening

Some AI-generated voices still carry audible traces. None of these signs is conclusive on its own, and a real person on a bad connection can show several of them, but they are worth noticing:

Flat or oddly placed emotion. The words say "I'm scared", but the tone stays even, or the stress lands on the wrong word.
Unnatural breathing. No breaths at all over long sentences, or breaths that sound pasted in at regular intervals.
Strange rhythm. Pauses in odd places, words that run together, or a pace that never changes.
Mispronounced names and places. Clones often stumble on family nicknames, local place names or words switched mid-sentence between languages, such as Hindi and English.
Too-clean or mismatched background. Studio-quiet audio from someone who claims to be at a roadside, or background noise that cuts in and out with the voice.
Metallic or warbling edges on certain sounds, especially "s" and "sh".

Why your ears are not enough

Here is the honest problem: listening by ear is unreliable. Good clones can sound natural. WhatsApp voice notes and phone calls are compressed, which removes much of the fine detail where artefacts would show up, and it also makes real voices sound slightly robotic. Stress plays a part too. When a message says someone you love is in trouble, you are not listening like an audio engineer.

So treat these signs as reasons to slow down, never as reasons to feel reassured. A voice that sounds perfectly normal can still be cloned.

The clues that matter more than the sound

Scam voice notes tend to follow a script. Look at what the message is asking you to do:

Urgency: "Right now", "before tonight", "don't wait".
Secrecy: "Don't tell Mummy", "keep this between us".
A new number or account: "My phone broke, this is my friend's number."
Money in hard-to-reverse forms: UPI to an unknown ID, gift cards, crypto or wire transfers. The FTC alert above lists wire transfers, cryptocurrency and gift cards as warning signs.
Requests for OTPs, PINs or passwords. No genuine relative, bank or official needs your OTP or UPI PIN.
Avoiding a live conversation: only voice notes, never a call you can return.

If two or more of these appear together, treat the message as suspicious whatever the voice sounds like.

How to check if a voice is AI: a step-by-step checklist

Pause. Urgency is the scammer's main tool. A few minutes of checking rarely costs anything real.
Do not reply with money, OTPs or PINs while you check.
Call the person back on a number you already knownot the one that sent the message. If you cannot reach them, call another family member or a friend who would know.
Ask something only they would knowor use a family code word agreed in advance (see our guide to family emergency voice cloning scams).
Get a second opinion on the audio. Run the voice note through a detector such as SpotAClone.
Report suspected fraud. In India, report suspicious calls and messages on Sanchar Saathi's Chakshu facility. If money has already gone, call 1930 or report at cybercrime.gov.in straight away.

Got a suspicious voice note? Check it free: 3 checks a day, no signup.

Check a voice note

What a detection tool can and cannot tell you

SpotAClone analyses the audio for patterns that tend to appear in AI-generated speech. It is built for Indian languages over phone-quality audio and accepts WhatsApp voice notes (.opus or .ogg) as well as .m4a, .mp3, .wav and other common formats. Audio is analysed and then deleted immediately; it is not stored. If you want to understand the method, read how voice deepfake detection works.

You will get one of three results:

Likely AI: the audio shows signs of AI generation. Treat the message as suspicious and verify before doing anything.
Likely real: no strong signs of AI generation were found. This does not prove who is speaking. A real human voice can still be a scammer, and detectors can miss a clone.
Unsure: the clip is too short, noisy or compressed to call either way. We would rather tell you that than guess.

Any detector, including ours, can be wrong. That is why we describe it as decision support, not proof. The result is one more piece of information alongside the context of the message and, above all, your callback.

Getting a more useful result

Upload the original voice note rather than a re-recording played through a speaker.
Longer clips with clear speech give the tool more to work with than a two-second fragment.
If the result is Unsure, do not re-upload until you get the answer you hoped for. Move on to verification.

Build habits that work even when detection fails

The strongest protection does not depend on spotting the fake. It depends on routines that a scammer cannot easily get around:

Agree a family code word for real emergencies.
Make "I'll call you back" a normal, unoffended response in your family.
Never share OTPs, PINs or UPI PINs. A bank will never ask for them. Read more in our guide to fake bank OTP calls with cloned voices.
Be aware of the related digital arrest scamwhere callers pose as police or officials.
If something has already gone wrong, see what to do if you sent money to a scammer.

The question "is this voice AI?" is worth asking. The question "have I confirmed this is really them?" matters more.

Sources

Frequently asked questions

Can I tell if a voice is AI just by listening?

Sometimes, but not reliably. Modern voice clones can sound natural, and phone or WhatsApp compression hides many of the small flaws you might otherwise notice. Treat what you hear as one clue, not as a verdict.

How do I check if a voice message is AI?

Combine three things: listen for warning signs, look at what the message is asking you to do, and verify the sender through a separate channel, such as calling them back on a number you already have. A detection tool like SpotAClone can add a second opinion on the audio itself.

If a detector says a voice is likely real, does that prove who is speaking?

No. A likely real result only means the tool did not find strong signs of AI generation. A real human voice can still belong to a scammer, and no detector is right every time. Always verify the person separately before acting.

Why does SpotAClone sometimes say Unsure?

Short, noisy or heavily compressed clips may not contain enough clear evidence either way. Rather than guess, the tool says Unsure. In that case, rely on verification steps such as a callback or a family code word.

What should I do if I think a voice note is a scam?

Do not send money, OTPs or PINs. Contact the person on a number you already know. In India, if you have lost money, call the National Cyber Crime Helpline 1930 or report at cybercrime.gov.in as quickly as possible.