Home · Use cases · Newsrooms

Verify audio before you publish.

Updated July 2026

Verify suspicious audio before you publish. A citable AI-voice verdict with the source model named in under half a second, backed by a published methodology.

VerificationEditorial review
Confidence
reported
Citation
Permanent URL
To verify audio in a newsroom, save the original file and run it through a detector before you publish. It returns the probability the voice is AI-generated, a confidence level, and the source model where recognized, with a published methodology and a permanent citation URL. It analyzes recordings, not live broadcasts, and it is one input to your reporting, not a replacement for sourcing.
The problem

Fake audio moves faster than a correction

Fabricated "leaked" recordings and cloned-voice clips of public figures now circulate hours before anyone can respond, and the 2024 cycle was the first where AI-generated robocalls reached voters at scale, serious enough that the US FCC confirmed AI-generated voices in robocalls are illegal. For a newsroom, publishing or amplifying a synthetic clip is a correction, a trust hit, and sometimes a legal exposure. Our deepfakes in elections note covers the pattern.

How newsrooms use it

Verify before you publish, cite after

Drop a suspect clip into the detector and get a probability, a confidence level, and the source model in under half a second. The methodology is published and version-stamped, so a verdict holds up in editorial review and can be cited in the piece with a permanent URL. When confidence is low on a compressed or short clip, the detector says so rather than guessing, which is exactly the honesty an editor needs. For the step-by-step, see how to verify AI audio.

Honest scope

A tool, not a verdict on the story

We analyze the audio, not the images, and not live broadcasts. We return whether the voice bears the signature of synthesis, which is one input to your reporting, not a substitute for it. Pair a verdict with your own sourcing, and use provenance standards like C2PA where available. Build checks into your workflow with the API. Used well, a verdict turns a viral rumour into a checkable claim your desk can stand behind, fast enough to matter on deadline and documented enough to survive scrutiny afterward.

Before you run it

Catch the obvious tells first

A quick listen narrows what to escalate. Unnaturally even pacing, missing or too-regular breaths, a suspiciously clean background, and a narrow emotional range are common signs a clip is worth verifying. None is proof on its own, which is why the detector reads the acoustic fingerprint the ear cannot. See AI voice vs human voice for the full checklist, and if the clip points to a scam or fraud angle, the fraud detection use case covers the response for security teams.

By the numbers

What every verdict gives you

0.48s
Median verdict
99%
Accuracy on clean audio
24+
Generators named
24h
Audio deleted after
FAQ · Newsrooms

Common questions

Can I cite the verdict in a story?
Yes. Every saved verdict has a permanent URL and an APA-style citation, and the methodology is published and versioned, so an editor or an opposing party can check the test against the exact method used.
How fast is it on deadline?
About half a second for a typical clip. Reading the full verdict, confidence and any model attribution, takes a few seconds more.
What if the clip is low quality?
Compressed or very short clips lower confidence, and the detector reports that rather than guessing. Treat a low-confidence result as inconclusive and seek a better source.
Does it work on video?
It reads the audio track of a video file, but it does not analyze the images. For a synthetic face, a video-specific tool is needed.
Is it free for journalists?
A single verdict is free with no account. Higher volumes and the API are on the paid plans.

Verify it before you rely on it.

Free for a single verdict. API for teams on Starter and above.

Open detectorSee pricing