Updated October 2026 · Voice rights, consent and deepfake labels
No single test reliably spots a deepfake voice, because generated speech keeps improving. Check the source, search for the same claim in trusted outlets, look for a disclosure, and be most careful when a voice asks for money or urgent action. Treat audio cues as clues, not proof.
Generated voices are getting harder to tell apart from real ones, and many listeners now hear a voice online that they cannot quickly judge. The temptation is to rely on a single sign, such as an odd pause or a flat tone. That approach fails, because the sign can be missing or can be faked. A better habit is to check the context, the source and the claim, and to slow down when the stakes are high.
This guide is for listeners, not for experts. It explains the checks that help, the ones that mislead, and what to do when you are unsure. It does not offer a guaranteed test, because none exists.
Ask who published the audio and how they came to have it. A verified account on a platform you trust is a better starting point than an anonymous page or a message forwarded from a stranger. Check whether the channel has a history of posting original material, and whether it links to a named person or organisation that you can contact.
Be careful with recordings that arrive without context. A clip with no date, no location and no speaker name should raise your guard, especially if it is shared with strong emotion and urgency.
A real event usually leaves traces. Search for the key claim, the quoted words or the event, and see whether reputable outlets, official sources or the people named have reported it. If the claim appears only in one place, and nobody else can confirm it, treat it with caution. Reverse searches for images and clips can also show where a recording first appeared.
When a voice seems to say something surprising about a public figure, check the figure’s own channels. Official statements are often the fastest way to confirm or refute a claim.
Some listeners notice unnatural breathing, odd emphasis, a flat emotional tone, repeated phrasing or a room sound that does not match the setting. These cues can be useful clues. They are not proof. A real recording can sound odd on a poor line, and a generated voice can be made to sound natural. Use the cues to decide where to look next, not to settle the question.
Notice whether the audio sounds too clean for the setting. A voice recorded outdoors with no wind, traffic or echo is worth checking, although many real recordings are also clean.
Creators who use generated voices are increasingly expected to disclose them, as covered in the disclosure post in this series. A disclosure near the start, or a written note in the description, is a good sign. The absence of a disclosure does not prove a deepfake, because many people publish without one, but it is a reason to check further.
Consider the stakes. A voice that asks you to send money, share a password, make a payment or take an urgent action deserves the most care. Scammers use urgency to stop you checking. If the message pushes you to act now and tell no one, that pressure is itself a warning. Pause, check, and contact the person on a number you already know.
For calls from people you know, agree a family or team code word that is never written down. A real relative will know it. Always call back on a known number before any payment, even when the voice sounds exactly right. The scam calls post explains these steps in more detail.
Some tools claim to detect generated speech. Their results vary, they can be wrong in both directions, and they may not work on compressed or re-recorded audio. Do not rely on a single score, and do not treat a tool as the final judge. Use it as one more input alongside the source, the context and a search for the claim.
If you publish audio, make it easy for people to check. Keep the raw recordings, publish a clear source note for each episode, and label any generated voice. Consistent provenance makes your real recordings easier to trust, and it makes fakes that copy your show easier to spot.
These checks reduce risk but do not remove it. A convincing fake can pass many tests, and a real recording can fail some of them. The safest habit is to verify before you act, especially when money or private information is involved.
A clean, documented recording is easier to trust than one that has been heavily processed without a record. The Vocal Checker shows clipping, noise and level problems in a file you own. It does not detect generated speech, so do not read its results as proof of authenticity. For a clean phone recording, the Tunedzy Go page covers the steps in your phone’s browser.
Spot clipping, noise, rumble and level problems. Runs in your browser, nothing is uploaded.
No single test is reliable. Combine checks on the source, the context and the claim.
No. They are clues at most, and real recordings can sound odd too.
Stop, call the person back on a number you know, and check with a family or team code word.
Use them as one input only. Their results vary and they can be wrong.
