LUNAROPS · OPERATIONAL UPLINK 100% UPTIME 1,247d POSTS 893 JEFF.MOON@LUNAROPS.DEV UTC --:--:--

What Mastering Engineers Actually Do

16 min read Audio Engineering
audio-engineeringmasteringloudnesslufsmusic-productionstreaming
Contents

Ask someone outside the music business what mastering is and you will usually hear some version of “making the song louder” or “the final polish.” Both are wrong in the way that matters. Mastering is the last stage of music production, the step between a finished mix and a released record, and its actual job is translation and quality control: take a stereo mix that sounds good in the studio it was made in, and make it sound good everywhere — on earbuds, in a car, on a phone speaker, on a club system — while meeting the technical specifications of every format it will ship to. For roughly two decades that quiet, careful craft was hijacked by a single destructive obsession with loudness, and the most interesting thing about modern mastering is that the technology of streaming has largely ended that war and handed the craft back to its real purpose. Understanding what mastering engineers do means understanding both the work itself and the loudness war that warped it.

The confusion is understandable because mastering is nearly invisible when done well. A mixing engineer’s work is audible — you can hear the vocal sitting forward, the drums punching through. A mastering engineer’s work is the absence of problems you would otherwise notice: the album that does not get harsh on the third track, the bass that does not disappear on your laptop, the loudness that is consistent from song to song, the master that does not distort when the streaming service re-encodes it. Mastering is the engineering discipline of making a recording robust to the brutal variety of the world it will be played in, and the people who do it well are doing something far more specific and technical than “polish.”

The mix/master boundary: two different jobs on the same audio

The cleanest way to understand mastering is by contrast with mixing, because the two are constantly conflated and they are genuinely different jobs operating at different scales. Mixing works inside the song: the mixing engineer has the multitrack session — dozens or hundreds of individual tracks, every drum mic, every vocal take, every synth layer — and balances them against each other, setting levels, panning, EQ, reverb, and effects on each element until the song has the right internal balance and depth. The mixer’s raw material is the separated parts.

Mastering works on the song as a finished whole. The mastering engineer typically receives a single stereo file — the printed mix — and can no longer touch the individual elements; they cannot turn the vocal up or the snare down in isolation. Instead they treat the entire mix as one object and make it better as a unit: adjusting overall tonal balance, controlling overall dynamics, optimizing loudness, and — critically for an album — making a dozen separately-mixed songs sound like they belong together, with consistent level and tone and the right amount of silence and crossfade between them. The mixer makes the song; the mastering engineer makes the record.

   MIXING                                MASTERING
   (inside the song)                     (the finished whole)

   [ kick ]---\                          single stereo mix
   [ snare]----\                              |
   [ bass ]-----+--> balance, pan,       tonal balance (broad EQ)
   [ vox  ]----/    EQ, reverb per       overall dynamics / loudness
   [ synth]---/     track -> STEREO MIX  stereo image, glue
                          |              album consistency + spacing
                          +------------> dither + format delivery -> RELEASE

This boundary also explains the recurring industry advice that a great master cannot rescue a poor mix. Because the mastering engineer is working on the blended whole, the things a bad mix gets wrong — a vocal buried under a guitar, a muddy low end where bass and kick fight — are baked in and cannot be cleanly separated out at the stereo stage. Mastering can nudge, glue, and polish; it cannot un-bake. The standing joke that “we’ll fix it in mastering” is, like “we’ll fix it in the mix,” usually a sign that someone is deferring a problem to a stage that cannot actually fix it. (Stem mastering, where the mixer delivers a handful of grouped sub-mixes instead of one file, is the compromise that gives the mastering engineer a little more leverage without handing back the whole session.)

What the work actually consists of

Strip away the mythology and a mastering session is a sequence of small, deliberate, technically-grounded decisions, almost all of them subtractive and subtle. The dominant activity is critical listening in an acoustically treated room on monitors the engineer knows intimately — the entire value proposition rests on hearing accurately what the mix engineer, working in a less-controlled space on the same speakers for weeks, could no longer hear objectively. The “fresh ears in a calibrated room” is not a marketing line; it is the core of the service, and it is why room acoustics and reference monitoring matter more in mastering than anywhere else in the chain.

On top of that listening, the tools are a small, focused set:

  • Tonal balance (EQ). Broad, gentle equalization across the whole mix — taming a harsh upper-midrange, adding a touch of air, controlling a boomy low end — usually in fractions of a decibel, often with both surgical digital and “musical” analog equalizers. The goal is a balance that translates across playback systems, frequently checked against reference tracks in the same genre.
  • Dynamics (compression and limiting). Gentle compression to “glue” the mix and control its dynamic range, and a limiter at the end to set the maximum level. This is the stage the loudness war turned into a weapon, but used correctly it is about controlling peaks and shaping dynamics, not maximizing volume.
  • Stereo image. Adjusting width, sometimes using mid/side processing to widen the high frequencies while keeping the bass mono and centered (essential for vinyl and for mono playback on phone speakers).
  • Loudness and metering. Setting the final integrated loudness and true-peak ceiling to hit the targets that matter for the delivery formats — the subject of the next two sections.
  • Sequencing, spacing, and fades. For an album, ordering the tracks, setting the gaps and crossfades between them, and matching their levels and tone so the record plays as a coherent journey.
  • Dither and format conversion. Converting from the high-resolution working format to each delivery format, applying dither when reducing bit depth to avoid quantization distortion.

Much of this leans on the same digital signal processing fundamentals that run through the whole audio chain — equalizers, dynamics processors, and loudness meters are DSP — but applied with a restraint that distinguishes mastering from mixing. The mastering engineer who reaches for 6 dB of anything is usually solving a mix problem that should have been sent back, not mastered around.

The loudness war: how “louder” hijacked the craft

For about twenty years, roughly the mid-1990s through the early 2010s, mastering was distorted by a phenomenon called the loudness war, and you cannot understand modern mastering without understanding what went wrong. The root cause is a quirk of perception: when two versions of the same audio are played back-to-back, the louder one almost always sounds better on first listen — fuller, more exciting, more “finished” — even though the loudness adds no real quality. This perceptual bias is robust and well-documented, and it created a destructive incentive.

In the CD era, every recording had a hard ceiling: 0 dBFS, the maximum digital level, above which samples simply clip. There was no playback normalization — whatever level you put on the disc was the level the listener heard, and a louder CD sounded more impressive than the quieter one next to it in the changer or on the radio. So labels and artists pushed mastering engineers to make records louder than the competition, and the only way to get more loudness out of a fixed ceiling is to reduce the dynamic range: crush the peaks down with aggressive brickwall limiting so the average level can be raised closer to the ceiling. Each record got a little louder, which moved the goalposts, which made the next record push harder — a textbook arms race where everyone’s loudness rose and nobody gained any relative advantage, while the music’s dynamics were progressively destroyed.

The cost was real and audible. Dynamic range — the difference between the quiet and loud parts that gives music its life and impact — was flattened. Transients (the snap of a snare, the pluck of a string) were squashed. In the worst cases the limiting was pushed so hard the audio clipped and distorted: the notorious example is Metallica’s 2008 Death Magnetic, so aggressively mastered that the distortion became a public controversy, with fans noting the game-version mixes (which bypassed the mastering) sounded cleaner. The TT Dynamic Range (DR) meter emerged as a way to quantify the damage, scoring releases on a dynamic-range scale and revealing how compressed modern records had become compared to their 1980s counterparts. A war that no listener asked for had degraded a generation of recordings.

How loudness normalization ended the war

The thing that ended the loudness war was not a change of heart in the industry; it was a change in how playback works, and it came from streaming. The key technology is loudness normalization: instead of playing every track at whatever level it was mastered to, the streaming service measures each track’s loudness and automatically adjusts playback so everything plays at roughly the same perceived level. The instant that happens, the entire incentive structure of the loudness war collapses — making your track louder no longer makes it sound louder to the listener, because the service just turns it back down to match everything else.

The measurement that makes this work is LUFS (Loudness Units Full Scale), standardized in ITU-R BS.1770 and the broadcast spec EBU R128. Unlike a simple peak meter, LUFS approximates how humans actually perceive loudness, using a frequency weighting (K-weighting) and a gating algorithm that ignores silent passages, producing an integrated loudness value for the whole track that correlates with how loud it subjectively seems. Streaming services pick a target integrated loudness and normalize every track to it:

Platform Normalization target (integrated) Notes
Spotify ~ -14 LUFS On by default; “Loud” setting ~ -11, “Quiet” ~ -19
Apple Music ~ -16 LUFS “Sound Check” normalization
YouTube / YouTube Music ~ -14 LUFS Only turns loud tracks down, not quiet ones up
Tidal ~ -14 LUFS Album and track modes
Amazon Music ~ -14 LUFS Comparable target
Broadcast TV (EBU R128 / ATSC A/85) -23 LUFS / -24 LKFS Much quieter; legally mandated in some regions

The practical consequence for mastering is profound. If you crush a track to -8 LUFS to win the loudness war, Spotify turns it down by 6 dB to hit -14 — and now your over-compressed, dynamically-dead master plays at the same volume as a track with healthy dynamics, except yours sounds flat and lifeless by comparison while the dynamic one sounds punchy. Normalization inverts the old incentive: loudness for its own sake now actively hurts you, because you paid for it in dynamics and got no volume advantage. The smart modern target is to master for good dynamics and a true-peak ceiling (commonly -1 dBTP to leave headroom for lossy-codec encoding overshoot), landing somewhere around the platform targets, and to stop chasing maximum loudness entirely.

The honest caveat is that the war is over but not erased. Normalization is on by default on the major streaming platforms, but not everywhere and not always — some users turn it off, some platforms historically did not normalize, club and radio playback may not, and a downloaded file has whatever loudness was baked in. And loudness remains a legitimate artistic choice: aggressive limiting is part of the sound of some genres, and a deliberately loud, dense master can be the right call. What changed is that loudness is no longer a free competitive weapon — it is a trade-off with a real cost, which is exactly how it should have been all along.

Deliverables: why one song needs several different masters

The least glamorous and most technical part of mastering is producing the actual deliverable files, and the key insight outsiders miss is that different formats need different masters. A mastering engineer does not make one file and ship it everywhere; the optimal master for streaming is wrong for vinyl, and the master for CD is different again. This is where the “translation” framing pays off most concretely.

For streaming and digital, the deliverable is typically a high-resolution file (24-bit, 44.1 or 48 kHz, or higher) with a true-peak ceiling around -1 dBTP and loudness in the neighborhood of the platform targets. Apple’s Apple Digital Masters program (formerly Mastered for iTunes) specifies delivering 24-bit masters so Apple can do the lossy encoding from the highest-quality source, with guidance on true-peak headroom to avoid the encoder clipping. Each platform then derives its own lossy codec versions from that master, which is why leaving true-peak headroom matters: a master slammed to 0 dBFS will overshoot and distort once it is encoded to AAC or Opus.

For CD, the deliverable is 16-bit/44.1 kHz audio — which means converting down from the 24-bit working master and applying dither, a tiny amount of shaped noise added during bit-depth reduction to prevent the quantization distortion that would otherwise appear in quiet fades. The whole album is usually delivered as a DDP (Disc Description Protocol) image, the standard package replication plants use, including track markers, gaps, ISRC codes, and metadata.

For vinyl, the master is genuinely different because the medium has physical constraints that digital does not:

  • Bass must be mono and centered. Out-of-phase low frequencies make the cutting stylus (and the playback needle) jump out of the groove, so a vinyl pre-master sums the low end to mono.
  • No brickwall limiting. The loudness-war tricks that work on a CD cause distortion and tracking problems on vinyl; vinyl masters are typically more dynamic by necessity.
  • High frequencies and sibilance must be tamed. Excessive high-frequency energy can overheat the cutting head and cause distortion, so de-essing and HF control are more aggressive.
  • Level versus playing time. Louder cuts require wider grooves, which means less time per side; a long album must be cut quieter or split across more sides.
   ONE FINISHED MIX  -->  several distinct masters

        +--> STREAMING : 24-bit, ~ -1 dBTP, ~ -14 LUFS, codec headroom
        |
   MIX -+--> CD        : 16-bit/44.1, dithered, DDP image + ISRC
        |
        +--> VINYL     : mono bass, tamed HF, less limiting, side-length aware
        |
        +--> HI-RES    : 24-bit/96kHz+, minimal processing

Add to this the routine extras — instrumental versions, clean/explicit edits, radio edits, the correct metadata and ISRC codes for royalty tracking — and the deliverables stage alone is a real chunk of the job, and one where a single wrong specification (wrong sample rate, missing dither, clipped true peak) can ship a flawed product to millions.

Do you still need a mastering engineer?

The honest modern question is whether this craft still needs a dedicated specialist, because the alternatives have gotten real. AI mastering services (LANDR and others) will analyze a mix and apply EQ, compression, and limiting algorithmically for a few dollars in minutes, and bedroom mastering with affordable plugins and metering is entirely feasible. For a lot of music — demos, background tracks, artists on tight budgets, genres where the production is deliberately rough — these are perfectly adequate, and pretending otherwise is gatekeeping.

What the specialist still offers is threefold and genuinely valuable for music where it counts. First, the calibrated, treated room and trusted monitoring — the ability to hear accurately what is actually on the recording, which no plugin grants you if your own room is lying to you. Second, fresh, objective ears: someone who has not heard the song four hundred times during mixing and can judge it as a listener will. Third, experience and accountability for the deliverables — knowing the format specs cold, catching the phase problem that will ruin the vinyl cut, making the album cohere. An AI service can apply competent processing; it cannot tell you the third track’s vocal is harsh because the mix needs revision, sequence your album for emotional flow, or take the blame when the DDP is wrong. The trade-off is cost and turnaround against that judgment, and where you land depends honestly on what the music is for. The same logic runs through the rest of audio production, from headphone amps and DACs on the monitoring side to live sound on the performance side: the tools have democratized, and the expertise has not become worthless — it has become a choice with clearer trade-offs.

Verdict

Mastering is not making things loud, and the most important development in the craft is that the technology of streaming finally made that obvious. The real work is translation and quality control: taking a finished mix and making it sound good across every playback system while meeting the technical specifications of every format it ships to. It sits at a different scale from mixing — operating on the finished stereo whole rather than the separated parts — which is why a great master cannot rescue a poor mix and why album consistency, sequencing, and format delivery are as much the job as EQ and dynamics. For two decades the loudness war perverted all of this into a destructive arms race that flattened dynamics for a perceptual edge that vanished the moment two tracks were normalized to the same level.

Loudness normalization ended that war by inverting its incentive: with Spotify, Apple Music, YouTube, and the rest measuring LUFS and turning every track to a common level, mastering for maximum loudness now costs you dynamics and buys you nothing. The craft that remains is the one that should have been there all along — accurate listening in a treated room, gentle and subtractive processing, and the unglamorous discipline of shipping the right master for streaming, CD, and vinyl, each with its own constraints. Whether that justifies a dedicated mastering engineer or an algorithm depends on what the music needs, but the principle does not change: mastering is the engineering of a recording’s robustness to the real world, and the moment it becomes about winning a volume contest, it has stopped doing its actual job.

Sources

Comments