Someone asks: “Is this real?”Without provenance, your word is all you have.
Before you publish, a mark goes into your official audio and its acoustic features are recorded. When a clip surfaces, you can show whether it came from the registered source, and whether it differs from the original.
Free pilot ／ Reply within 3 business days
For Japan-based projects only. Inquiries in English are welcome.
We check authenticity only: the registered source, and any differences from the original. We do not judge whether what is said is true, or whether the speaker is who they appear to be.
official-announcement.wav
- 01 Before releaseRegistered by
- Communications Office, Example City
- Three layers of provenance
- Audio watermarkEmbedded in the audio
- Audio fingerprintAcoustic features registered
- C2PA signatureSIGNED SOUNDtime-stamped
- 02 After releaseSocial media
- A reposted clip matches the registered original, no differences foundChecked by watermark and fingerprint
- 03 When asked“Is this real?”
- Answer with the verification result, not just your word
Illustrative screen. A verification result shows matches and differences against the registered data. It does not judge whether the content is true.
Six places this problem shows up
Any of these sound familiar?
An evacuation order or a mayor’s press conference is shared around, and someone asks if it is genuine.
Details for local governments (Japanese) →A clip of something you broadcast spreads with the caption “did they really say that?”
Someone clones your chief executive’s voice.
Audio you published is reused somewhere you never authorized.
A recording arrives, supposedly from a colleague, and you cannot check it.
Your podcast is clipped, and you cannot tell what was cut.
All six have the same starting point.
Tell us yoursYou can’t detect your way out of this.
Detection gives a probability, after the clip has spread. Provenance has to be recorded before the audio leaves your hands.
- Chases each new generation method
- Misses clips that were cut or spliced, not synthesized
- Responds only after the clip spreads
When the stakes are high, “probably fake” settles nothing.
- Marks your audio and records its features before release
- Clips can often still be matched to the original
- Shows the registered source and any differences
The provenance exists before the incident does.
Use both. Detection checks other people’s content. Provenance marks your own.
True stories took about six times as long as false ones to reach 1,500 people on Twitter (Vosoughi et al., Science, 2018; news in general, not audio).
Register it before you need it.
One step before release, so you are not reacting from behind.
Mark it before you publish
A mark goes into the original and its features are recorded. The effect on sound quality is kept low.
Check the clip against it
A circulating clip is compared with the original. Copies that lost the mark can often still be matched.
See where it came from
The result shows the registered source and any differences.
A watermark and a fingerprint, both built for audio
Matching works from the sound itself, not from metadata a re-upload can strip. *Heavy re-encoding or processing can defeat matching.
A reference point that predates distribution
The moment you register becomes your anchor for checking later clips.
No special equipment, no new recording kit
Use any recorder and any channel. The mark is added after recording.
It does not tell you whether what is said is true. It answers one question: did this audio come from the source that registered it, and does it differ from what was registered? It does not judge whether the speaker is who they appear to be, and it does not stop fakes from being made.
What changes when the clip starts spreading
It does not take disinformation down, and it does not judge whether content is true.
| Situation | Before | After |
|---|---|---|
| Someone asks: “Is this real?” | All you can do is point at the official page | You show the registered source and any differences, grounded in a verification result |
| Your audio spreads on social media | You cannot be sure whether it is yours or an edit | You can often match it to the original and see any differences |
| What it covers | – | Provenance and authenticity only |
Can your audience check the source?
Listeners can use the trust mark as a starting point to check the source. “Unregistered” does not mean “fake”: it means nobody registered it.
Under the defined verification conditions, no differences from the registered data were found.
The audio differs from the registered data. The cause needs a separate look.
Nobody registered this content.
*Illustrative. Display depends on the channel and implementation. No dedicated app is expected.
Who is behind this
Technology partner
Watermarking and fingerprinting by Evixar Inc. Its EAW and EFP are on the C2PA Soft-Binding Algorithm List (com.evixar.eaw.1 / com.evixar.efp.type3.1).*A listing of the algorithms, not C2PA certification or membership.
Field trials
Under a program of Japan’s Ministry of Internal Affairs and Communications, a proof-of-concept with Kyoto Prefecture on official audio. No production reference cases yet.
Operator
Shachihata, the maker of Japan’s name seals, supplies third-party reference information for checking provenance.*Reference information only. Not a judgment on truth, and no statutory presumption, including under Japan’s Act on Electronic Signatures and Certification Business (Act No. 102 of 2000).
Start with a conversation, not a contract.
The pilot is free. Production is quoted per project.
Frequently asked questions
What is SIGNED SOUND?
Does it also tell me whether what is being said is true?
Is it admissible as evidence in court? Does it have legal effect?
How is this different from a deepfake detector?
What is the “free pilot”?
Where is the audio data stored?
Do we have to change how we publish today?
Record the provenance before you need it.
Ask for a conversation or a free pilot. For Japan-based projects only; inquiries in English are welcome.
(1) This service verifies only integrity (sameness with the registered data) and provenance, from the point of registration onward. It does not prove that what is said is true, and it does not prove who the speaker is. The source that registered the audio is not necessarily the speaker.
(2) Use of this service does not guarantee admissibility as evidence in court or any specific legal effect.