Upload a reference track. Reveal the production blueprint behind it.
Analyze your track →One takes someone else’s record apart. The other three measure the file you made yourself — and those are free.
Is it ready to master? A checklist with a verdict: headroom, loudness, dynamics, mud, the low end in mono, phase.
No reference needed. Only the faults that are facts and not taste: true peak over the ceiling, clipping already baked in, a fold to mono that goes wrong.
A scorecard side by side with a reference you choose, and a score you can take apart rather than a verdict you have to believe.
Real sections from a real analysis — not a mock-up of a dashboard you'll never see.
The drum stem goes against fingerprints of the classic boxes — 808, 909, LinnDrum, DMX and the rest of the family tree. Not "vintage-style drums": the machine, and what gave it away.
Per melodic layer: the synthesis approach, the filter movement, the character. When a patch could come from three machines, it says so and ranks them — a guess you can check beats false certainty.
The moves that make a record sound finished, read from where they live in the audio — ready to recreate in any DAW.
Where the track actually sits, so you can place it, reference it and write in the same world.
You did not come for a list of gear. You came because a record sounds like something you want to reach. The analysis exists to make that path checkable: what the music is built from, how it was put together, and how firm the ground is under each part of the answer.
We work on the most energetic ninety seconds of the track. Averaging a six-minute arrangement smooths away the very things that identify a machine — the attack, the filter moving, the tail of a decay. Detail survives in a section; it dies in a summary.
That section is separated into its layers before anything is measured, so every number afterwards describes one instrument at a time rather than the sum of them. A kick measured through a full mix is mostly a measurement of everything else.
Each layer is described in signal terms before anything tries to name it: attack and decay envelopes, spectral centre and roll-off, harmonic against percussive balance, inharmonicity, onset density, tonality and grid. These are facts about the audio, and they do not change if the naming step is wrong.
Each separated layer is matched against fingerprints of real machines using a learned audio representation — the same layer, compared like with like. Only then does the naming step run, and it does not get to invent: it receives the measurements and the matches, and it is bound by them.
A drum-machine match taken from a separated stem is strong enough to be binding: the name it produces is preserved exactly, even when our own reference notes have nothing to add about that machine. A synthesiser match on the same path is real evidence but not binding — it is weighed against the measurements. Anything below the similarity threshold is treated as a prior, not as a finding. The report is built on that hierarchy, in that order.
Every identification carries a confidence, and when a patch could plausibly come from more than one machine the report ranks the alternatives instead of choosing one. Swapping the machine the evidence points at for a more famous one that is easier to write about is the most damaging error available here, so it is ruled out by construction. Where the evidence is thin, the report says so in words, not in silence.
Naming the gear is the means, not the point. What you are meant to leave with is a description of how the sound was made that you can take back to your own session and try — the synthesis approach, the movement, the treatment, the space. If you get closer to the record you were chasing, the analysis did its job, whatever the plate on the front of the machine said.
Plugins that imitate hardware convincingly can read as the hardware, because at the level of the signal they were built to. A layer buried under heavy bus processing is harder to call than an exposed one, and the confidence reflects it. Vocals are analysed and never exported. None of this is guesswork dressed as certainty: when we are unsure, that is what you get told.
Ninety seconds of it comes back taken apart — the machines, the synthesis and the moves behind it, with the evidence for every call.
Analyze your track →3 analyses free · no card
By creating an account you accept the Terms of Use.
Every analysis you have run is still here
One link, good for an hour
We'll send a link to set a new password. It works once and expires in an hour.