Home › Guides › Loudness compensation
Loudness compensation: mixing at low volume without losing the bass
Turn a mix down by 20 dB and you do not hear the same mix, quieter. You hear a thinner one. Here is why, what the ISO 226 equal-loudness contours actually ask for, and how to compensate without making things worse.
- Your ear loses bass faster than midrange as the level drops. It is physiology, measured since Fletcher and Munson in 1933 and standardised as ISO 226.
- The right correction is not the equal-loudness contour itself but the difference between two contours: the one at your reference level and the one where you listen now. At the reference it is zero.
- In the bass that difference comes to about 0.6 dB of boost for every dB you turn down. It belongs on the monitoring path, never on the master bus.
Why a quiet mix sounds thin
A microphone has one frequency response. The ear has a different one at every level. In the cochlea, amplification is strongly compressive at the high-frequency end and much less so at the low-frequency end, so perceived loudness grows faster with level in the bass than in the midrange. Turn the volume down and the bass falls away faster than everything else.
The effect is large. Going from a very quiet 20 phon to a loud 90 phon takes, by definition, 70 dB at 1 kHz. At 31.5 Hz the same step in perceived loudness takes only 38.9 dB. The bass has about half the dynamic range of the midrange, so every decibel you take away costs it roughly twice as much in perceived level.
That is why quiet mixes often do not translate. You push the low end up to hear it, and on a louder system the mix turns boomy.
From Fletcher-Munson to ISO 226
Fletcher and Munson published the first equal-loudness curves from Bell Labs in 1933. Robinson and Dadson re-measured them in 1956, and their curves became ISO 226:1987, later found to carry large errors below 1 kHz. A new model by Suzuki and Takeshima, built on measurements from several laboratories, became ISO 226:2003.
The standard was revised again as ISO 226:2023. Its authors put the differences from the 2003 edition at 0.6 dB at most and consider the two practically the same. When producers talk about “the Fletcher-Munson curve” today, the measured, modern version is ISO 226.
The loudness button gets it wrong
The loudness button on a hi-fi amplifier applies an equal-loudness contour as a boost, often a fixed one. That is a category error. The contour describes your own ear, and the engineer who balanced the mix listened through that same ear at the reference level: the contour is already in the balance. Apply it again and you apply it twice.
The numbers show how far off it is. At 20 Hz the 80 phon contour sits 39 dB above its own 1 kHz value, so a control that applies the contour boosts 20 Hz by 39 dB at the reference itself, where nothing should change. For a 30 dB drop, the correct correction asks for 15.8 dB. Holman and Kampmann made the point to the Audio Engineering Society in 1978: the right basis is the difference between contours.
The right correction: the difference of two contours
Call LR the level at which the mix was balanced and Δ how far you turned down. Every frequency drops by Δ, but the ear does not need the same drop everywhere to keep the same balance. The correction is the difference between the two contours, normalised at 1 kHz:
It is zero at the reference by construction, a bass boost that grows as you turn down, and a bass cut above the reference. Fierro, Rämö and Välimäki derive the same expression in their 2019 work on adaptive loudness compensation.
One more property makes it practical: the curve depends almost only on how far you turned down, not on the absolute level. For references between 75 and 90 phon, the curves for a given drop agree within 0.06 dB at −10 dB, 0.16 dB at −20 and 0.34 dB at −30.
How much bass, in numbers
Between 20 Hz and 2 kHz the difference of contours is very close to a single low shelf. The table gives the shelf gain that holds the balance of an 83 dB SPL reference as you turn down.
| Volume | ≈ dB SPL | Bass shelf | What that level is |
|---|---|---|---|
| 0 dB | 83 | 0.0 dB | the film, broadcast and mastering reference |
| −5 dB | 78 | +3.0 dB | a comfortable sustained level in a treated room |
| −10 dB | 73 | +6.0 dB | where much small-room work happens |
| −15 dB | 68 | +8.9 dB | quiet, attentive listening |
| −20 dB | 63 | +11.8 dB | about conversation level at one metre |
| −30 dB | 53 | +17.3 dB | late-night checking |
The whole behaviour fits in one number: about 0.6 dB of bass for every dB you turn down.
What level should you mix at?
The published monitoring references sit between 82 and 85 dB SPL. ITU-R BS.1116-3 aligns each loudspeaker to 85 − 10 log n dB(A), which gives 82 dB(A) per loudspeaker for a stereo pair. SMPTE RP 200 uses −20 dBFS pink noise at 85 dB(C) per channel. Bob Katz's K-20 and K-14 use 83 dB(C).
Those levels are where hearing is closest to linear, and they are loud for a long session in a small room. Many engineers work lower. Compensation is what lets you do that without your bass decisions drifting.
Do you need to calibrate?
No. Because the correction depends on how far you turned down rather than on the absolute level, it only needs a truthful zero: the volume position at which you actually make your balance decisions. Leave the volume where you normally work and call that your reference.
Getting the absolute level wrong costs little: between a 75 and a 90 phon reference the curve moves by at most 0.34 dB. Calibration still earns its place. It gives you a readout in dB SPL, comparability with the standards, and the certainty that your “normal” level is not 95 dB. Our monitor calibration guide walks through the procedure.
What compensation cannot do
- The contours come from pure tones, heard in a free field by young listeners with normal hearing. Music is broadband and your ears are not the standard's average, which is why an amount control matters: 100 % is a starting point, not a target.
- A bass shelf ignores the top octave. At −30 dB, ISO 226 also asks for about +2 dB at 10 kHz and +5 dB at 12.5 kHz. The data up there are sparse and high-frequency hearing varies a lot with age, so most compensators leave it alone. It is a choice, not an absence of error.
- It restores tonal balance, not everything else. Distortion, the room's contribution, spatial impression and masking still change with level.
- A commercial reference track was balanced by someone else at an unknown level, so no compensator can know its true reference. Set its amount by ear against your own mix.
Compensation is not normalisation
Loudness normalisation matches the playback level of different programmes: a streaming service does it to its whole catalogue. Loudness compensation corrects how your hearing perceives the frequency balance at a given level. They solve different problems and you may need both: matched levels for a fair comparison, explained in our guide to level-matched A/B against a reference track, and compensation for working quietly.
Put it on the monitoring path, not the master bus
A compensation plugin on the master bus shapes what you hear, but also what you export if you forget to bypass it. It does not know your listening level either: you have to move its setting every time you touch the monitor volume. Plugins exist, such as LSP Loudness Compensator, which offers both the 2003 and 2023 contours, or APU Loudness Contour.
The natural place is after everything that gets printed: the monitoring path, driven by the monitor volume itself. A monitor controller can do that; a plugin in the session cannot.
How PHON does it
PHON is our studio monitor controller for Mac. Its loudness compensation follows the method above:
- one low shelf per output at 90 Hz, within 0.75 dB of the exact ISO 226 difference from 20 Hz to 2 kHz at −20 dB, and 1.24 dB at −30;
- its gain read from the volume, so nothing is applied at 0 dB and it fades out as you come back up;
- nothing above the reference by default, nothing above 2 kHz;
- an amount in percent, stored separately for your mix and for the reference source.
Next: A/B your mix against a reference track, level-matched
The details of PHON's implementation are in chapter 6 of the user manual (PDF).
References
- H. Fletcher and W. A. Munson, “Loudness, its definition, measurement and calculation”, J. Acoust. Soc. Am. 5(2):82–108, 1933.
- D. W. Robinson and R. S. Dadson, “A re-determination of the equal-loudness relations for pure tones”, Br. J. Appl. Phys. 7:166–181, 1956.
- Y. Suzuki and H. Takeshima, “Equal-loudness-level contours for pure tones”, J. Acoust. Soc. Am. 116(2):918–933, 2004.
- Y. Suzuki, H. Takeshima and K. Kurakata, “Revision of ISO 226 from the 2003 to the 2023 edition: the background and results”, Acoust. Sci. & Tech. 45(1), 2024.
- T. Holman and F. Kampmann, “Loudness compensation: use and abuse”, J. Audio Eng. Soc. 26(7/8):526–536, 1978.
- L. Fierro, J. Rämö and V. Välimäki, “Adaptive loudness compensation in music listening”, Proc. 16th Sound and Music Computing Conference, Málaga, 2019.
- B. Katz, Mastering Audio: The Art and the Science, 2nd ed., Focal Press, 2014.
- ITU-R BS.1116-3, Methods for the subjective assessment of small impairments in audio systems, 2015.