Veriframe verifies recordings the same way it verifies video and images: against the indexed original, never against a guess. Publish a speech, interview, or broadcast once — anyone holding a clip can check whether it still matches, and if it doesn’t, the verdict names the exact seconds where it diverges.
Why it matters
The recording is checked in windows of a few seconds, and one divergent window is enough to break authentication. Averaging over the whole track would hide a replaced sentence; per-window comparison finds it — and names the seconds.
A thirty-second quote cut from an hour-long recording aligns back to its exact position in the original, even when the cut lands mid-sentence. The full context is one click away.
Re-encoding, resampling, and volume changes by platforms and podcast apps keep the recording anchored to its original — reported as re-encoded, never as tampering.
How VERA works for audio
Your team publishes the original recording to the Veriframe registry at release time. The registry stores its acoustic fingerprints in overlapping windows of a few seconds each — not just the file.
A journalist, fact-checker, or listener drops the clip they received into the public verify page. No account needed.
The clip is aligned to the original by its loudness contour, then every few-second window is compared separately. One divergent window is enough to break authentication — and the verdict says which seconds diverged.
What the verdict tells you
Every verdict states its evidence. These are the outcomes the audio pipeline distinguishes today:
Platform recompression, format conversion, resampling, and volume changes keep the recording anchored to its original — reported as re-encoded, not as tampering.
A fragment cut from a longer recording is aligned back to its position in the original — even when the cut lands mid-sentence.
A sentence replaced mid-recording lights up its exact window while the rest still matches. The verdict localizes the divergence to the second.
A recording that shares nothing with the indexed original is exactly that — different audio. VERA reports "not found" rather than speculating about what it might be.
Use cases
Index the official recording when it is released. When a clip with rearranged words starts circulating, the verdict shows exactly which seconds diverge from what was actually said.
Publish the episode and index it in the same step. Every quote clip in circulation can then be traced back to its place in the conversation — trimmed is fine, spliced is named.
A contested clip takes moments to check against the registry, and the verdict — with the diverging seconds named — has a permalink you can cite.
Honest limits
Veriframe proves whether a recording matches an indexed original. It does not — and does not pretend to — listen to an arbitrary clip and rule whether the voice is cloned. That claim cannot be made reliably, so we never make it.
If the original was never indexed, VERA answers "not found" — honestly. The registry model is the point: proof, not probability.
A copy buried under added noise may fail to match its original. The failure direction is deliberate: a degraded copy returns "not found" — never a false confirmation.
The public verify page accepts MP3, WAV, M4A, OGG and FLAC alongside video and images.