Which measurement?
Name the quantity. If it cannot be named, there is nothing underneath.
What the score is, and what it isn't.
What AudioVibes AI measures, and where the measuring stops. Part of ABA Media Platform.
You have been handed a number by a machine before. The trouble is not that those numbers are wrong.
It is that measurement and opinion come back in the same font, at the same size, with the same confidence, and nothing on the screen tells you which is which.
So you discount all of it, which is the sensible move and also a waste of a perfectly good measurement.
AudioVibes AI gives your track a number and a letter too. Here is exactly how that is made, and where it stops being trustworthy.
The solid part
The measurements are not opinions.
Loudness, loudness range, true peak, stereo correlation, spectral tilt, clipping, DC offset. These are computed to published international standards — ITU-R BS.1770-4 and EBU R128 — and they are not opinions. Two correct implementations of BS.1770 measuring the same file agree, and ours is checked against an independent implementation, pyloudnorm, to keep it that way.
If the tool tells you your integrated loudness is −11.3 LUFS and your true peak is −0.98 dBTP, those are facts about your file. Take them to the bank.
The soft part
The score is a different kind of thing.
The score is those measurements weighed against a profile — expectations for what a well-made record in your genre tends to look like. Rock expects a different loudness range than classical. Cinematic expects a different spectral tilt than hip-hop.
Those profiles are declared heuristics. That means we wrote them down from experience and published them, and it means nobody has yet measured a body of real commercial releases to check whether they are right.
So the score is a considered opinion with the reasoning shown. It is not a measurement, and we will not describe it as one.
The condition, not the promise
What would change that.
A reference corpus. Real commercial masters, per genre, measured with the same engine on the same standards, so each profile's expectations come from evidence rather than from our judgment.
That work is underway. It is unglamorous — largely buying second-hand CDs and ripping them — and it is the only thing that turns a declared heuristic into a calibrated one.
When a genre's profile clears that bar, the label on it changes and the score for that genre starts meaning something stronger. Until then it does not, and this page will keep saying so. We would rather publish the condition than a promise.
An instrument you can keep
How to use the score today.
Here is a test you can run on any tool in this category, this one included. Take one finding and ask three questions:
What target?
The value it was compared against, stated as a number.
Where did the target come from?
A published standard, or somebody's judgment. Both are legitimate. Not knowing which is not.
A measurement answers with a published standard. An opinion answers with somebody's judgment. A tool that cannot answer at all is asking to be trusted, which is a different transaction than the one you came for.
Every finding here names the measurement behind it, the target it was compared against, and why the difference matters. If you disagree with the target, you know exactly which number to argue with — and that argument is a better use of your time than the score itself.
The findings are the product. The score is a summary of the findings.
What happens to your audio
Stated in the words that were approved.
Your audio never leaves your machine. It is decoded and analyzed in your browser, and it is never transmitted.
The measurements do travel: every completed report is handed to the page that embeds the tool, and for signed-in members the host stores it against their account. The audio is not part of that — the measurements and the file name, nothing more.
Both are quoted verbatim, and the second never appears without the first. The rulebook carries the whole file, including what is still open.
Why publish this at all
Because the honest version is uncomfortable.
Every tool in this category could write this page. Most do not, because the honest version is a list of things the tool cannot yet support, and that is an uncomfortable thing to put next to a download button.
This one was written by the person who wrote the engine — one man, an AI team, no employees. There is nobody here to hide behind and no committee to blame for a claim that does not hold up.
A number you can interrogate is worth more than a number you have to trust, and the only way to prove we mean that is to show you where our own number gets weak.