What a voice recording report can and cannot prove

Our opinion first: a report on a recording that does not tell you what it cannot prove is not worth much. The people who read these reports are rarely the people who ordered them. A fraud manager, a claims handler, a lawyer, sometimes a judge. They deserve to know exactly how far the document goes.

This applies whatever produced the report: a forensic audio expert, an in-house investigator, or a software tool. Here is how we read one, and the questions we would ask of our own.

First question: which file, exactly?

A report is about one file. Not "the recording" in general, not the version someone forwarded later. A good report ties itself to that file so that nobody can quietly swap it.

The usual way is a cryptographic fingerprint, often written as SHA-256. It is a long string computed from every byte of the file. Change one byte and the fingerprint changes completely. If someone later produces "the same recording", anyone can compute its fingerprint and compare. Same string, same file. Different string, different file, whatever it sounds like.

A report without a fingerprint, a date and a clear description of the file is weaker than it looks.

Second question: what does it actually conclude?

Read the conclusion word by word. "Consistent with", "no indication of", "inconclusive" and "proves" are four very different claims, and serious reports pick their words with care. Watch for a confident sentence in the summary that the rest of the report does not support.

"Inconclusive" deserves respect. Some recordings simply do not carry enough to judge. A report that never says so is guessing, and a guess dressed as a finding does more damage than an honest "we cannot tell".

What a report on a recording cannot prove

• Who is speaking. A question about the recording is a different question from which person it is.
• Consent. Whether the person knew they were recorded, or agreed to it, is a matter of facts and law, not of the file.
• Legal validity. Admissibility and weight belong to the judge and the rules of the jurisdiction.
• The next clip. A conclusion on one recording says nothing certain about another recording of the same voice.

Where a report helps

Triage, mostly. A team that receives dozens of recordings a week cannot send each one to an expert. A structured first report helps them pick the few that deserve that cost. It is also a written record: on this date, this exact file was examined, and this was the result. Months later, in a dispute, that record is worth more than anyone's memory of the call.

It helps the honest side too. When copied voices are common, a genuine recording becomes easy to wave away with "that could be AI". A clear report on the right file moves the discussion from suspicion back to evidence.

How to check a recording and keep a report you can use

1. Start from the original file. Not a forward, not a screen recording. The fingerprint only means something if it is the file you were given.
2. Write down where it came from. Who sent it, when, on which app or line.
3. Compute or record the fingerprint before anything else. Then store the original untouched and work on copies.
4. Keep the report with the file. Store both together, unchanged, and quote the report's limits whenever you pass it on.

Plans are on the pricing page.

FAQ

Is a report on a recording proof in court?

No report decides a case on its own. It can support a decision or a request for an expert, but weight and admissibility belong to the court.

Why would a report say inconclusive?

Because some recordings do not carry enough to judge. An honest inconclusive is more useful than a confident guess.

What is a SHA-256 fingerprint for?

It identifies the exact file that was examined. If someone later presents the same recording, its fingerprint shows whether it is the identical file or a different one.