The ear has an EQ curve, and the level knob is wired to it

Take a 1 kHz tone at some level. Play a tone at another frequency and ask a listener to adjust it until it sounds equally loud. Repeat across the audible range and the result is one contour; repeat at a different 1 kHz level and the result is a different contour, not a shifted copy of the first. The loudness level of any point on a contour is named in phon, equal by definition to the SPL of the 1 kHz tone it was matched against.

Fletcher and Munson published the first set in 1933 with headphones and a handful of listeners; Robinson and Dadson repeated the work in free field in 1956 and theirs became ISO 226 in 1987. A multinational re-measurement led by Suzuki and Takeshima found the low-frequency part of Robinson and Dadson's curves to be substantially in error, and ISO 226:2003 replaced them. The 2023 edition adjusts that by at most 0.6 dB; its authors describe it as the same standard for practical purposes. The 2003 parameters, which are the ones widely enough reproduced to be checked, are what is computed here.

Two features of the shape do all the work. The curves are steep at the bottom, because the ear is insensitive to low frequencies at low levels. And the curves flatten as they rise: at 90 phon the ear needs far less help at 50 Hz than it does at 40 phon. That second feature is the one with consequences.

Read the 50 Hz row. On the 40 phon contour a 50 Hz tone needs about 38 dB more than 1 kHz to sound as loud; on the 90 phon contour it needs about 18. A balance that was right at one level has the bottom octave 20 dB out at the other, and nothing in the signal chain changed.

• A mix built loud and played quiet sounds thin. At soundcheck volume the contour flattered the bottom end; at background level it does not. The reverse — built quiet, played loud — is boomy. This is what the "loudness" button on domestic amplifiers tried to correct, and why it never quite worked: the correct curve depends on the actual level at the ear, which the amplifier does not know.
• The same is true across the audience. The front row and the back are on different contours, before any air absorption is counted. The spreading loss that costs the back row 20 dB also moves it onto a curve that needs more bass, and the low end is the part of the system least able to provide it there.
• Sensitivity is not the same at every frequency, and the peak is not at 1 kHz. The dip in the contours around 3–4 kHz is the ear canal's resonance: at any level, that region is heard louder than the meter says, which is why it is where harshness lives and where a couple of decibels of cut goes a long way.
• Ten phon is roughly twice as loud. The sone scale that sits on top of the phon doubles every 10 phon above 40. A system that is "twice as loud" is 10 dB up at 1 kHz, and — because the contours converge — rather less than that in the bass.

If the contours say how the balance shifts with level, they also say how to shift it back. Take the contour at the level the material was balanced at, take the contour at the level it is being reproduced at, and the difference between them — referred to 1 kHz — is the EQ that restores the balance:

G(f) = (ref − cur) − [ Lp(f, ref) − Lp(f, cur) ]

At 1 kHz the contour passes through the loudness level by definition, soG(1 kHz) is identically zero. The curve can only ever change tonal balance, never level — that is arithmetic, not a design choice, and it is what keeps the idea from being a slow automatic fader.

Contourtonist is this arithmetic as a plugin. It measures how loud the room actually is — from a calibrated measurement microphone in its standalone, or from any SPL meter's live output — and applies the curve for that level on a live output, fitted onto sixteen minimum-phase biquads so it adds no latency. When a noise limit pulls a show down 8 dB, the bass does not leave with it. There is a browser demo running the same DSP compiled to WebAssembly, with an on-screen fader standing in for the room, and a user guide.

A-weighting is the meter's attempt to hear like an ear. Its curve is loosely described as the 40 phon contour inverted, and the table shows how loose: it tracks the contour's shape through the midrange and diverges at both ends, because it was fitted for convenience as much as for fidelity. It is nevertheless what nearly every noise regulation is written in, so a limit of 100 dB(A) is a limit on a spectrum that has had 19 dB taken off it at 100 Hz and 39 at 31.5 Hz.

C-weighting is nearly flat from 31.5 Hz to 8 kHz and is the closer match to a contour at concert level. The gap between a C reading and an A reading of the same signal is therefore a measure of how much low-frequency energy is present: for a spectrum with equal level in every octave from 31.5 Hz to 8 kHz the gap is 1.8 dB, and a real bass-heavy show can open it wider. A licence that sets a dB(A) limit and a dB(C) limit a fixed amount above it is limiting the bass separately, and that is usually the intent.