Home › Voice Detection › For Podcasters

Your voice, AI narration, and disclosure.

Creators come to voice detection from three different directions: checking whether submitted voice work is human, working out what has to be disclosed when AI narration is used, and dealing with someone cloning their own voice. There is also a fourth thing nobody warns you about — well-produced audio is disproportionately likely to be flagged as synthetic when it is not. All four are covered here.

Check an audio clip free Why good audio gets flagged ↓
3 checks a day, no signup Audio never stored
The trap

Why your well-produced audio gets flagged.

This is the part that catches creators out, so it goes first.

Detection leans on the small irregularities that mark audio as physically captured: room tone that persists through silences, audible breath, uneven energy inside syllables, high-frequency detail that nothing tidied up. Podcast production removes almost all of it, deliberately.

What you do in postWhat it removes
Noise reduction / gatingThe room tone and noise floor that prove a microphone was in a space
Compression and limitingThe natural unevenness of energy across a delivery
De-essing and EQHigh-frequency character detection reads as physical
Breath removalOne of the strongest single human cues, edited out on purpose
A treated boothRoom reflections — a clean space genuinely sounds unnaturally clean
The consequence is unfair and worth naming: a voice artist working to broadcast standard in a treated booth is more likely to be wrongly flagged than someone talking into a phone in a kitchen. If a platform, client or listener challenges your audio, submit the rawest take you have rather than the published master, and point them at our page on false positives.
Disclosure

What has to be disclosed.

Rules are tightening in several directions at once, and the practical answer is simpler than the legal one.

Three separate regimes can apply to the same episode, and they are not the same thing.

  • Platform policies. Several major platforms now require creators to label synthetic or substantially AI-generated media at upload, and enforcement is via the platform rather than a regulator. This is the one most likely to affect you first.
  • Statutory transparency rules. The EU AI Act's transparency provisions are the most-cited example of a legal requirement to disclose AI-generated content; other jurisdictions are moving in similar directions at different speeds.
  • Advertising and sponsorship rules. If synthetic voice appears in an ad or a sponsored segment, advertising standards and endorsement rules may apply independently — particularly if a synthetic voice implies a real person's endorsement.

Whether any of these binds you specifically is a question for your own legal counsel, not for a guide. But the practical answer is straightforward: disclose it. A line in the show notes and a sentence in the episode costs nothing, satisfies platform requirements, pre-empts the statutory direction of travel, and removes a category of problem entirely. Creators who disclose AI narration rarely regret it; the reverse is not true.

Guest audio is worth a line in your release form. If a guest submits pre-recorded audio, an explicit statement about whether it was synthetically generated or voice-cloned turns a future argument into a contract term. Cheaper than any detector.
Your own voice

If someone clones your voice.

Podcasters are unusually exposed: hours of clean, well-recorded speech in a single voice is close to ideal source material, and it is published for anyone to download.

  1. Capture the evidence before it moves. Save the file, the URL, the account that posted it, and the date. Screenshots of the posting context matter as much as the audio.
  2. Report to the platform. Most major platforms have a synthetic-media or impersonation reporting route distinct from generic abuse reporting. Use the specific one — it routes to people who understand the claim.
  3. Report to the generating provider if you can identify it. Cloning a real person without consent breaches the terms of every major synthesis service, and they have abuse channels for exactly this. Note that identifying the provider from audio is generally not possible — our detector does not attribute — so this applies when the poster has said which tool they used.
  4. Get legal advice on the right claim. Depending on where you are, personality or publicity rights, passing off, defamation and data-protection rules may all be in play. Which one applies is jurisdiction-specific and worth an hour of a lawyer's time.
  5. Tell your audience. A short, calm post from the real account is often the fastest practical remedy, and it inoculates your listeners against the next one.
Not forensic-grade. A TextSight voice result is a triage signal, not proof, and it must never be the sole basis for a disciplinary, employment, financial or legal decision. Our full position on what a result is worth is here.
FAQ

Creator questions.

Why was my professionally recorded podcast flagged as AI?
Because production removes the evidence of physical capture. Noise reduction, compression, de-essing, breath removal and a treated booth all strip out the irregularities detection relies on. It is a known and genuinely unfair pattern in this category: people who record well are more exposed to false positives than people recording on a phone.
What should I submit if my audio is challenged?
The rawest take you have, not the published master — ideally an unprocessed recording straight from the interface, plus outtakes and alternate reads. Rendered synthetic narration has no earlier version, whereas a real session leaves a trail of files behind it. That trail is stronger evidence than any detector score.
Do I have to disclose AI narration on my podcast?
Platform policies, statutory transparency rules such as the EU AI Act's provisions, and advertising standards can each apply, and whether any binds you specifically is a question for your own legal counsel. Practically, disclosing costs nothing and removes the problem — a line in the show notes and a sentence in the episode.
Can I check whether a submitted voiceover is human?
You can run a check, and you should read it as one signal rather than a finding. The stronger move is to ask the contributor for raw session files and outtakes, which a genuine recording produces and a rendered voiceover does not. Building that into your contributor agreement is cheaper than checking after the fact.
Someone cloned my voice. What can I actually do?
Preserve the evidence, report through the platform's specific synthetic-media or impersonation route, report to the generating provider if the poster identified it, get jurisdiction-specific legal advice on which claim applies, and post a calm correction from your real account. The last one is often the fastest practical remedy.
Does publishing my podcast make cloning easier?
Realistically, yes — hours of clean single-speaker audio is good source material and it is public by design. That is not a reason to stop publishing. It is a reason to agree a family safe word, be sceptical of urgent voice requests that claim to be from you, and have a plan for reporting an impersonation quickly if it happens.
Related

More voice detection guides.

How accurate is it, really?

Why we publish no single accuracy percentage, what false positives look like, and how to weigh a result.

See our position →

Choosing a detector

The questions to ask any vendor, the claims to distrust, and how to run your own test.

Read the buyer's guide →

Compare two clips

The control-recording method: run a known-genuine clip alongside the suspect one and read the gap.

Learn the method →

Checking a voiceover? Read the caveats first.

Three checks a day, free, no signup. Your audio is never stored.

Check an audio clip All voice guides
Mastered audio is more exposed to false positives, not less