Punchy, dynamic, compressed — three words, two measurements

Why a record can be punchy and flat at once, and why none of these words can be turned into a setting.

The tonal words have a clean story: three regions, one threshold, and a difference between two spectra at the end of it. The dynamics words do not, and the mess is worth understanding because it is the mess itself that carries the information.

There are two measurements underneath six words, and they measure different things at different timescales.

Crest factor: the millisecond

Peak against average, in decibels. How far the loudest instant stands above the general level of the file.

That is a question about transients — the leading edge of a snare, the pluck of a string. A record where those edges survive has a high crest factor. A record where a limiter has taken the tops off has a low one, no matter how the rest of it behaves.

Loudness range: the minute

LRA, in loudness units. The spread between the quiet passages and the loud ones, measured over the whole track.

That is a question about arrangement and about slow gain movement — whether the record goes anywhere. A song with a whispered verse and a wall-of-sound chorus has a high LRA. A song that sits at one level for four minutes has a low one, even if every drum hit inside it is razor sharp.

They disagree, and that is the point

Nothing forces these two to move together, and on this library they visibly do not. Four records are punchy or better on crest while measuring consistent on range:

Nation of Millions is the clearest case. A crest factor of 18.1 dB puts it above almost everything in the collection for transient life — those drums hit as hard as anything recorded. And an LRA of 3.4 puts it among the flattest: the record never lets up, from the first bar to the last.

Both facts are true. Both are audible. "Dynamic" as a single word cannot hold them, and that is why there are two measurements and not one.

A record can hit hard and go nowhere. Sample-based hip-hop does it on purpose; the loudness war did it by accident.

For contrast, the two ends of the collection:

• ·Dire Straits, Brothers in Arms* (1985)** — crest 22.4, LRA 10.6. High on both. A 1985 all-digital recording with the transients intact and the arrangement allowed to breathe.
• ·Red Hot Chili Peppers, Californication* (1999)** — crest 8.9, LRA 3.8. Low on both, and the narrowest record in the library at 0.117. The canonical loudness-war casualty, and it measures the part.

The space words

Two more, on the simplest measurement here: side energy against mid energy, one ratio for the whole file.

A Love Supreme (1965) is the widest at 0.731 — a jazz quartet spread across a room by two microphones and left alone. Californication is the narrowest at 0.117.

One caveat that matters more than it looks: this single ratio hides how width changes with frequency. Nearly every record is narrow at the bottom and wide at the top, and a record with an unusual shape there — mono lows with an extremely wide top, say — measures the same as one that is evenly moderate. The per-band figures exist for that reason and are what catches a fake-stereo transfer.

Now the part that surprises people

None of these six words becomes a chain parameter. Not one.

The tonal words end up in an EQ curve, computed as a difference. The dynamics words end up nowhere — and the reason is the same reason mastering stays a craft.

A measurement of a finished record tells you the result. Compression is a process, and several different processes produce the same result. An LRA of 4 LU is equally consistent with:

• ·a fast attack at a low ratio, catching everything gently, or
• ·a slow attack at a high ratio, letting every transient through and then clamping hard behind it.

Those two sound nothing alike. One sits still; the other punches. And the finished file cannot tell you which one happened, because the information that would distinguish them was consumed in making it.

So when Wavely derives a preset from a record, the compression ratio is marked inferred — modelled from crest, LRA and loudness spread together, and flagged as a plausible guess rather than a reading. The attack and release times are inherited from the genre base, untouched, because there is nothing in the measurement that could set them honestly.

Which means the answer to "how do I get that record's punch" is not in the numbers. It is in the two compressors, and in deciding for yourself.

Why keep the words at all

If they set nothing, they earn their place two other ways.

They are how a record is found. The matcher reads text, so punchy and consistent are how someone reaches Nation of Millions without naming it.

They are how a preset gets caught lying. A record tagged one thing and measuring its opposite is a preset that will be reached by the wrong requests — and since the tags are derived from the same measurements the chain uses, the disagreement is detectable. The library has been wrong about itself twice, both times because a reputation outlived the transfer it was formed on.