THE LAB / EXPERIMENT 02 · MASTERING

One mix.
Three very different masters.

We built one stereo mix from 33 separate recorded parts (called stems), then mastered that exact recording three ways. Hear how processing changes the music's loud-and-quiet contrasts, from a more natural version to an extreme example of the “loudness war,” where making music consistently loud can cost it some punch and expression.

Same original performance40-second comparisonSame tone adjustments in all threeReal audio, measured waveforms
ACTUAL 40-SECOND WAVEFORMSL + R · same scale
DynamicDynamic: measured stereo peak and RMS waveform
ModerateModerate: measured stereo peak and RMS waveform
BrickwalledBrickwalled: measured stereo peak and RMS waveform

Measured from the original published 40-second masters: positive and negative peaks plus average (RMS) level, left and right channels. Tap any waveform to open its detailed full-size view. These images require no audio downloads; local replacement audio doesn't change the pictured published masters.

Start at a comfortable volume. The loudness-war sample is intentionally heavily compressed and distorted. This is a teaching example, not a recommendation for how music should sound. The original-loudness versions are very different in level.

01 / HEAR THE DIFFERENCE

Hear it, then compare it fairly.

Change versions at the same point in the song. Try the loudness versions first, then repeat at equal volume to hear what the processing itself changed.

Before you listen: Choose a test and press Play. Switching between versions keeps you near the same point, but may pause briefly. Want to compare an original album release with a later remaster? Open Optional: replace online audio with local WAV files below the player to add your own versions of the same song. Your files stay on your device. Local playback does not guarantee pause-free switching.

The loudness progression

Original mastered levels

Listen in order. A is most dynamic, B is louder and more controlled, and C deliberately pushes the music hard into brickwall territory.

Ready to load your local audio
0:00 / 0:40
0:000:100:200:300:40

Choose a version and press Play. Waveform images above require no audio downloads.

Optional: replace online audio with local WAV files

Have an original release and a later remaster of the same song? Put one in A and the other in B to hear what changed. You can also compare your own mixes. Choose WAV files from your device; nothing is uploaded.

To compare an original master with a remaster, replace both A and B. Any slot you don't replace still plays our example. Choose the same song at the same starting point, and match the listening volume yourself. The waveforms and measurements below still show our original examples.

Already have files from our demo? These batch options recognize the demo filenames:

Published examples play online. Any files you choose manually stay on your device.

03 / THE REAL RESULTS

What the measurements actually show.

The full-song numbers below come from the final MAGDA WAVs. LUFS estimates how loud a complete song seems on average: a value closer to zero generally means louder playback. Crest factor is the gap between the highest sample and the song's average signal energy: a bigger gap can leave more contrast between sharp hits and the background. dBTP is a peak-level estimate that accounts for peaks between digital samples. These numbers explain what changed, not which master is better.

A / Dynamic

−20.98

LUFS · full-song average loudness

Crest factor20.27 dB
Est. true peak−3.27 dBTP

B / Moderate

−13.83

LUFS · full-song average loudness

Crest factor14.04 dB
Est. true peak−1.37 dBTP

C / Brickwalled

−8.48

LUFS · full-song average loudness

Crest factor5.51 dB
Est. true peak−1.55 dBTP

SAME 40 SECONDS / 2:30–3:10

Peak-to-average contrast

Dynamic
19.04 dB
Moderate
12.94 dB
Brickwalled
5.22 dB

The same passage loses 6.11 dB of crest factor from A to B and another 7.71 dB from B to C. These are measured changes in the completed stereo audio, not guessed settings.

LUFS measures program loudness. The dBTP values are four-times-oversampled estimates of true peak; MAGDA's native limiter was not established as a dedicated true-peak limiter. The final versions were measured and trimmed to keep estimated reconstructed peaks under −1 dBTP.

04 / BEHIND THE MASTER

Exactly what we changed, and why.

Each version started from the same entire 3:13.7 stereo rough mix, built from 33 aligned untreated Raw stems of “Diva.” We never changed individual instrument levels, panning, performance or the triangle between these mastering versions.

01 · Raw stems33 untreated recordings, aligned in time
02 · One rough mixFixed levels and panning, no mix compression
03 · Same mastering EQIdentical broad tonal adjustments in A, B and C
04 · Different dynamicsMinimal → moderate → deliberately extreme
01 · EQ: what the three versions have in common

Before changing dynamics, we used the same three broad EQ moves in every mastered version. That keeps the main A/B/C comparison about compression, saturation and limiting rather than different bass or treble settings.

Low-shelf boost: +1.25 dB at 88 HzAdds a little low-end weight to bass and kick.
Bell cut: −1.85 dB at 320 HzReduces some low-mid thickness that can make instruments feel crowded.
High-shelf boost: +0.95 dB at 8,000 HzAdds a small amount of brightness to cymbals and the triangle.
All three filters use broad Q valuesQ = 0.75, 0.80 and 0.75; not surgical cuts.
What this can sound like

A somewhat lighter low-mid region, slightly fuller bass and a little more top-end air. To actually isolate EQ, use the EQ Only tab: both tracks there are matched to −23 LUFS, and no compressor or limiter is active.

02 · A / Dynamic: preserve the musical contrasts

We applied the shared EQ and a +3.5 dB fixed gain. The compressor, saturator and limiter were bypassed. No track-by-track adjustments were made.

Compression: bypassedNatural peak-to-average differences remain.
Hard clipping / saturation: bypassedNo intentional waveform flattening.
Limiting: bypassedTransients were not deliberately constrained by a brickwall limiter.
Measured output: −20.98 LUFSFull-song crest 20.27 dB, estimated true peak −3.27 dBTP.
What to listen for

The triangle's sharp attack and ringing, the solo's balance against the band, and the louder-versus-quieter notes in the ping-pong synth. This is our dynamic mastered reference, not the untouched rough mix: shared EQ still changes its tone and crest factor slightly.

03 · B / Moderate: compression and controlled limiting

With the same EQ as A, we increased the fixed input gain to +9 dB and engaged a stereo-linked compressor. A limiter then controlled the highest peaks.

Compressor threshold: −16 dBCompression begins around this signal level, subject to the processor's knee and detector.
Ratio: 2.3:1 · attack: 28 msAbove threshold, the compressor reins in rises; the slower attack allows some initial hit through.
Release: 175 ms · knee: 6 dBRecovery is gradual rather than an immediate snap.
Makeup: +3 dB · stereo linkedRaises post-compression level while keeping left and right treatment coordinated.
Limiter: −2.2 dB threshold1.5 ms attack; 140 ms release; final output trim −2.3 dB.
No added saturationThe saturator is bypassed. This is the middle treatment.
What changed in the measurements

Loudness rose from −20.98 to −13.83 LUFS. Full-song crest factor fell from 20.27 to 14.04 dB, while the same 40-second section fell from 19.04 to 12.94 dB. Listen for whether the guitar solo stands out more against the band and whether the louder and softer synth bounces begin to sound closer in level. The measured crest change is real; the particular instrument-balance judgment is a listening question, not a measured track-level change.

04 · C / Brickwalled: deliberately excessive loudness-war processing

We retained the identical EQ, then pushed the master much harder. This version combines heavy stereo compression, hard clipping and aggressive limiting. It is deliberately exaggerated for teaching, not a recommended commercial master.

Input gain: +12 dBPushes the same mix harder into the dynamics processors.
Compressor threshold: −21 dBMore of the song crosses the compression region.
Ratio: 7.5:1 · attack: 1 msMuch stronger reduction with a much faster reaction to transients.
Release: 130 ms · knee: 2 dBTighter, more forceful gain control than the moderate version.
Compressor makeup: +15 dBPushes the compressed signal into the next stage.
Hard saturator: +6 dB driveMAGDA Hard mode, 80% wet mix, output −1 dB. Intentionally creates harmonic distortion and clipped peaks.
Limiter threshold: −9 dB0.3 ms attack, 110 ms release, and −5.6 dB final output trim.
Estimated true peak: −1.55 dBTPThe final post-limiter trim avoids the positive reconstructed peaks seen in the early draft.
What changed in the measurements

The full master reached −8.48 LUFS with only 5.51 dB of crest factor. In our 40-second test passage, crest fell further to 5.22 dB. Listen for the guitar becoming more continually prominent and for stronger and quieter synth bounces losing some of their contrast. Some details may stand out more even as the recording becomes denser. These extreme settings also add clipping and limiting, so not every difference can be attributed to compression alone. Listen for distortion, pumping and masking as well as the musical trade-offs.

05 · Why fair comparison matters

The dynamic, moderate and brickwalled masters differ by more than 12 LU overall in their original output levels. Louder recordings can seem more impressive even when the dynamics processing makes them less enjoyable.

For the fair test, we took the exact same 40-second passage from each final master and changed each file's gain only until all three measured −23.00 LUFS. We did not re-compress, remix or EQ those comparisons. This lets you compare the character of the processing without the obvious loudness advantage.

Blind ABX is a test of discrimination, not preference

The optional ABX challenge asks whether an unknown recording is identical to reference A or B. It doesn't tell you which one you prefer. Training runs in your browser; scored games check your answers through our separate scoring service. Your listening files are not uploaded from your device.

06 · Why a remaster can sound like a remix

Have you ever heard a remaster where a guitar solo seems louder, backing vocals stand out or a synth comes forward? It can sound as though someone changed individual track levels, but that is not necessarily what happened.

Compression, limiting, EQ and other mastering processes can change how we hear the relationships inside a finished stereo mix. Here, the original instrument levels and panning are fixed. In the more heavily processed versions, the guitar may appear more prominent and the synth's louder and quieter bounces may converge. A more noticeable instrument does not necessarily retain more of its original expression.

The takeaway

Sometimes mastering makes you hear more of an instrument, but less of the performance within it. A remaster can sound like a different mix without actually being a remix. Stronger processing is not automatically better, and a small settings change can create an unexpectedly audible difference during a particular passage.

Our compressor responds to the combined stereo signal, not to each instrument crossing a separate threshold. The heaviest treatment also uses clipping and limiting, so its effects cannot all be assigned to compression alone.

07 · What our experiment can and cannot prove

It demonstrates: the measured progression of program loudness, crest factor and reconstructed true peaks, produced from exactly the same source mix and EQ. It can also help listeners investigate transient impact, perceived density, clarity and distortion.

It does not prove: that all commercial mastering sounds like these extreme examples, that any one version is necessarily best, or that a change in the triangle’s perceived prominence must be caused by the triangle itself. Mastering processes the whole stereo mix, so interactions with cymbals, guitars and vocals matter. The original supplied album master uses different mixing decisions and is not one of these controlled mastering comparisons.

Our approach keeps the source fixed, documents every processing decision and checks the exported files. The processing is deliberately exaggerated as an educational choice, and final musical and editorial judgment belongs to The Sound Question’s curator. Measurements inform those decisions; they do not replace listening.

SOURCE / CREDIT / METHODS

Recording credits and methods

Music: “Diva” by Tom McKenzie from Extraordinary. Source: AlbumDB by Thomas McKenzie, Eloi Moliner Juanpere and Alec Wright, listed under CC BY 4.0.

For more about the source recordings, visit The Recordings Behind Our Labs.

Research paper: McKenzie, Moliner and Wright, AlbumDB, AES 160th Convention (2026), Paper 10278.

Our changes: We built one stereo rough mix from 33 Raw stems, then independently mastered it three ways using MAGDA. These 40-second excerpts are educational adaptations, not Tom McKenzie's official masters. No endorsement or collaboration is implied. Audio streams from our host; optional files selected in your browser remain on your device.

All quoted measurements refer to the final 44.1 kHz, 24-bit stereo WAV exports. Details and file checks are saved in the project’s final MAGDA acceptance manifest. Public listening demonstration. No analytics or user-file uploads.