This is a specimen. It is not a real finding about a real recording.

TR‑8F29A1 does not exist. No audio was submitted, no analysis was run, and no person is described here. The document below is a filled‑in example of the format Truthring intends to issue, published so that a lawyer, an editor or a compliance reviewer can judge the form of it in advance. Every figure is either marked [VERIFY] or is an obvious placeholder. The page is excluded from search indexes, and it must not be circulated as evidence of anything.

Specimen document · Truthring verification report

Specimen verification report

What Truthring sells is not a percentage. It is a document that somebody who was not in the room can pick up, read, and check — including the parts that argue against the conclusion.

What a verification report is

A dated document tying one analysis to one file: the verdict and its confidence, the engine version that produced it, the signals that drove the score in plain language, the false positive rate for that class of audio, and the limits of what may be inferred. It is designed to be handed to a third party and disputed.

TR‑8F29A1 Truthring verification report · Lacewing Technologies

Specimen — not a real finding Issued [VERIFY: issue date — no report has been issued]

01Document

Report ID: TR‑8F29A1 illustrative
Ringmark: TR‑8F29A1 — the reference for this analysis. Quote it in any correspondence about the result.
Issued: [VERIFY: issue date not set]
Methodology version: [VERIFY: engine and methodology version not fixed] Method
Issued by: Lacewing Technologies, Navi Mumbai, Maharashtra, India
Status: Specimen. Not issued to any party. Carries no evidential weight.

02File examined

Every value in this block is an invented placeholder, present to show which properties a real report records. They describe no actual file.

File name: example‑clip.m4a illustrative
Container and codec: MPEG‑4 audio, AAC‑LC illustrative
Duration: 0:14 illustrative
Sample rate: 16 kHz, resampled from an 8 kHz source illustrative
Channels: 1 (mono) illustrative
File hash: SHA‑256 — [VERIFY: no file, no hash]
How it reached the submitter: Stated by the submitter as a voice note forwarded by a third party, not recorded by the submitter. Truthring cannot confirm a provenance account and does not attempt to. illustrative
Prior processing: Codec history indicates at least two compression passes before submission. Read the limitations section before treating this as a finding about the file’s origin. illustrative

03Assessment

Likely synthetic

Generation signatures present, at a recording quality sufficient to trust them

Confidence · illustrative

Illustrative wording. In a real report this paragraph states, in one or two sentences, what the engine found and what lowered the confidence. Here it would read: signatures consistent with a speech-generation system were present across the usable speech, and confidence was reduced by phone-line compression that removed part of the frequency range the analysis relies on.

Verdict: Likely synthetic · illustrative
Confidence: 87 · illustrative
Attributed generator: Not named in this specimen
Source quality: Phone‑compressed · 0:14 · illustrative
Usable speech: 11.2 s of 14.0 s · illustrative
Engine version: [VERIFY: not fixed]

The two verdicts are not symmetric, and a reader should know which one they are holding. Likely synthetic is the stronger answer, because reaching it requires positive evidence in the signal. Likely human is weaker than it sounds: it can mean the speech is genuine, or it can mean a codec destroyed the evidence somewhere between the speaker and the file. This specimen shows the stronger form. A real report says explicitly which situation applies.

04Signals observed

Written in plain language rather than feature names, because a report is useless to a recipient who cannot follow it. This describes the class of evidence, not the engine’s internals, and nothing proprietary is disclosed by it. The wording below is illustrative.

Band reconstructionEnergy above roughly 4 kHz showed the regular, banded structure characteristic of speech reassembled by a vocoder, rather than the uneven decay a microphone produces in a real room.
Prosodic rangeRhythm and pitch variation across the clip sat inside a narrower band than the reference distribution for natural conversational speech of comparable length. Narrow on its own means little; narrow alongside the band evidence means more.
Vocal-tract coherenceFormant behaviour remained consistent with one physical speaker throughout, so the clip shows no sign of having been assembled from several voices.
Splice boundariesNo discontinuities of the kind left by joining separately generated segments. This is an absence, and an absence is weak evidence in either direction.
Room and backgroundAmbient noise did not behave the way a real acoustic space behaves, though heavy compression makes this the least reliable of the observations here.
Codec historyConsistent with the submitter’s account of how the file travelled. This corroborates the account; it does not confirm it.

05False positive rate for this audio condition

The figure below is the rate at which the same engine calls genuine human speech synthetic, on audio of this class — not the headline rate across all conditions. It appears on the report itself because a verdict without it cannot be weighed, and because a reader should not have to go looking for the number that argues against the conclusion.

[VERIFY: false positive rate for telephone-quality audio not measured] Genuine human speech of this recording class, wrongly reported as synthetic. Measured on a held-out evaluation set. [VERIFY: evaluation not yet run; sample size unknown]

Read it as a population statistic. It tells you how the engine behaves across a test set of similar audio; it does not tell you the probability that this particular clip is one of the errors. If the consequence of being wrong is serious, the report is a screening step and not the decision. Both directions of error are published, by condition, on the accuracy page.

06Limitations of this result

Specific to this analysis, not a generic disclaimer.

Fourteen seconds is a short clip. Less speech means fewer independent observations, and confidence is lowered accordingly rather than reported as though the clip were longer.
The file was compressed at least twice. Compression removes exactly the high-frequency detail several of the observations above depend on. A cleaner copy of the same recording, if one exists, would produce a more reliable answer.
This says nothing about who made the clip, when, or why. A generation signature identifies a method, not an author, an intent or a date.
Nothing here authenticates the words spoken. A recording can be synthetic and accurate, or genuine and misleading. The analysis addresses how the audio was produced, not whether its content is true.
The provenance account is the submitter’s. Truthring examined a file. It did not witness how that file came to exist and cannot corroborate a chain of custody.
No result may be the sole basis for an adverse decision about a person. See compliance.

07Privacy: what happened to the audio

Audio: Erased at the completion of analysis. Not retained, not attached to this report, not used for training. [VERIFY: deletion behaviour not implemented; the product is pre-launch]
Derived features: Erased with the audio.
Retained: This verdict, the engine version, and a one‑way hash of the file, so the analysis can be matched to a file somebody else holds without our holding the recording.
Retention period: [VERIFY: retention period not set] Four objects, four lifetimes
Disclosure: No copy has been provided to any third party. Any lawful request received would be counted at transparency.

08How a recipient checks this report is genuine

If you have been handed this document by somebody else, you should be able to confirm it came from us and has not been altered, without needing the audio and without taking the sender’s word for it. That is the whole point of issuing a document rather than a screenshot of a score.

[VERIFY: verification mechanism not built] No route currently exists. Nothing on this specimen resolves, no code can be looked up, and no signature can be checked. What is intended is that a Ringmark can be entered somewhere on this site and returns the verdict, the confidence, the engine version and the issue date held in our record, so that a recipient can compare them against the document in their hand and see any discrepancy immediately.

Until that exists, treat any document in this format as unverifiable. A report you cannot check is worth no more than the sender’s assurance, which is the situation this product exists to improve on.

09Actions

These are inert on a specimen. Nothing downloads, nothing is shared, and no link is generated.

Disabled deliberately. Producing a downloadable file from a fabricated report would be the one thing on this page capable of doing real harm.

Specimen document. Truthring reports a probability, not proof, and this specimen reports nothing at all. A real report is issued by Lacewing Technologies, Navi Mumbai, Maharashtra, India, names the engine version that produced it, and carries the false positive rate for the audio condition assessed. No Truthring result should be the sole basis for accusing, disciplining, dismissing or prosecuting anyone.

Why the document, and not the number

A detection score handed over on its own puts the recipient in an impossible position. They are given a conclusion with no method attached, no error rate, no version, and no way to tell whether the person passing it on has quoted it accurately. Most disputes about synthetic audio are lost at exactly that point, because the party on the other side asks a reasonable question and nobody can answer it.

The report format exists to move the argument onto ground where it can be had properly. It states what was examined, what the engine measured, how often that engine is wrong in this direction on this kind of audio, and what the result specifically cannot establish. A well-prepared opponent should be able to attack it, and the sections above are written so that they can. A document that cannot be attacked has not disclosed enough.

The limitations section is the part we would most like people to read. It is longer than the assessment on purpose. If a report ever appears with a confident verdict and a short list of limits, something has gone wrong in how it was produced.

What is real on this page and what is not

Real: the structure, the section order, the commitment that a false positive rate for the specific audio condition appears on every report, and the commitment that limitations are stated for the individual result rather than as boilerplate.
Not real: the report ID, the file, the duration, the sample rate, the confidence figure, the signals, and the verdict. There is no recording. There is no analysis. There is no engine running behind this website.
Not held: any certification. Truthring has no SOC 2 report, no ISO 27001 certificate and no independent audit of any kind, and no report it issues will imply one.

If you have arrived here from a link somebody sent you, and they presented this as a finding about a recording, it is not one. Tell us and we will look into how it was circulated.

Questions about the report format

Is TR-8F29A1 a real analysis of a real recording?

No. It is a specimen: a filled-in example of the document Truthring intends to issue, published so that a lawyer, an editor or a procurement reviewer can see the format before committing to anything. No recording was analysed, no person appears in it, and the page is excluded from search indexes. Any figure on it is either marked as needing verification or is an obviously illustrative placeholder.

Why does the report state a false positive rate rather than just a score?

Because a confidence figure on its own cannot be acted on. What a reader needs to know is how often the same engine, on the same kind of audio, calls a genuine recording synthetic. A report that gives you a score without that denominator is asking to be believed rather than checked, and it is the reason detection results get misused.

Can a Truthring report prove that a recording is fake?

No. It records a probability produced by a stated method on a stated file, together with the error rate for that class of audio and the limits of the inference. It cannot establish who created a clip, when, or why, and it is not sufficient on its own for any decision that harms a person. Courts decide admissibility, not us.

How will a recipient check that a report is genuine?

A verification route is intended, so that somebody holding the document can confirm the Ringmark, the engine version and the verdict against the issuing record without needing the audio. The mechanism has not been built and nothing on this specimen resolves to anything.

Why is the audio not attached to the report?

Because keeping the recording is the part with real privacy cost, and the report is designed to survive without it. Truthring's commitment is that audio is erased once analysis ends and a one-way hash is kept in its place, so a file you hold can still be matched to the analysis. That trade is explained on the retention page.

Reports

The method behind it

Trust

Reviewed 31 August 2026