Mastering: multiband, Dynamic EQ, saturation, limiting, and normalisation.

Mastering turns a mix into a deliverable: a file with the level, balance, and compatibility required by each destination. The typical chain is not long, but each link has its own moment and failure. After overall EQ and compression come local-intervention and level tools: Multiband, Dynamic EQ, Saturation, Limiting; platform loudness normalisation (Streaming normalization) acts on them, and QC closes the process by deciding whether the file goes out or returns to the desk.

If mix balance and space are unfinished, complete them first: what the mixing guide resolves should not be resolved in mastering, and decisions in saturation, stereo, phase, and automation constrain what you can ask of the limiter.

Multiband

Multiband compression divides the spectrum into bands (usually three or four, crossing around 100-300 Hz and 3-5 kHz), applying independent dynamics processing to each. Its value is intervening in one region without touching the rest: tame bass surging in drops without squeezing vocals, lift high-frequency air only when needed, or contain 2-5 kHz so loud sections do not grate.

Its risk matches its advantage: band crossovers. Each filter adds phase around the crossover, each band’s release creates uncoordinated movement, and the typical result is pumping, a “rubbery” feel, or lost transient punch. Mastering therefore uses little margin: at most 2 to 4 dB reduction per band, gentle compression, and individual band bypass listening to confirm the problem still exists without processing.

Practical rule: if the problem occurs only at certain moments, multiband is too much firepower; if it affects the whole track consistently, it may make sense.

Dynamic EQ

Dynamic EQ combines an equaliser with a detector: the filter acts only when its band’s signal exceeds threshold. It handles problems that appear and disappear. A resonance troublesome only on low guitar notes, 3 kHz harshness when vocals shout, sibilance in strong esses: in each case, a narrow 2-to-4-dB cut happens when needed and withdraws as soon as the signal falls.

Compared with multiband, it works on one filter without extra crossovers, introducing less phase and greater transparency. Compared with serial compression, it leaves the track’s overall dynamics untouched. A de-esser, for example, is Dynamic EQ tuned to sibilance; a narrow-Q 4 kHz dynamic equaliser does the same job on an instrument.

Starting setting: narrow Q on the problem frequency, threshold just above the level where the problem is audible, fast attack so the cut arrives on time, and short release so it disappears with the passage. Then always compare bypassed at matched loudness.

Saturation in mastering

Mastering saturation seeks the same goal as mixing saturation but with less error margin: harmonic density and cohesion without raising peaks. A gentle layer of tape, valve, or console on the main bus makes the mix sound more “finished”, helps treble translate on small systems, and rounds harder transients before the limiter.

Order matters. Saturating before limiting rounds peaks and reduces limiter work; saturating afterwards adds peaks above your set ceiling and forces gain reduction. Measure before and after saturation: many processors raise peak levels and distort the measurement on which the remaining chain depends.

Excess is the trap. Saturation reduces dynamic range and softens transients, so aggressive main-bus drive produces a lower-peak but flatter-feeling master. Use small amounts, compensate gain, and volume-match A/B: if the saturated master sounds louder, the comparison is invalid.

Limiting

The limiter is the final ceiling guardian. It sets maximum peaks (usually -1.0 to -1.5 dBTP for encoder margin) and compresses anything attempting to exceed them, with short lookahead and program-dependent release to avoid audible modulation.

Amount is the critical decision. Between 2 and 4 dB reduction on the highest peaks supports natural sound; beyond that, punch is lost and background noise rises with makeup gain. If the desired level requires more, the problem is not the limiter: raise the vocal or hook in the mix, cut excess bass consuming margin, or accept a master with greater range.

A clipper before the limiter absorbs isolated peaks at a controlled distortion cost, reducing limiter work; this is common in density-focused genres. Always listen where inter-sample distortion appears: with a -1.0 dBTP ceiling and 320 kbps encoder, the encoded file may exceed 0 dBFS even when the WAV did not.

Streaming normalization

Loudness normalisation equalises volume between songs: the platform measures integrated loudness with the ITU algorithm and adjusts playback gain towards its target. Spotify’s target is around -14 integrated LUFS, Apple Sound Check starts lower around -16 LUFS, and YouTube operates near -14; exact figures change and should be confirmed in each service’s documentation.

The practical consequence is uncomfortable for those accustomed to hot mastering: mastering to -8 LUFS is no louder on the platform, merely less dynamic after normalisation. A dynamic master retains contrasts and sounds fuller after adjustment. The consistent approach is to choose the working level with normalisation in mind, check results with loudness meters, and leave True Peak margin.

The exception is a destination without normalisation: a club PA does not measure LUFS, and a hotter, more compressed master does translate into perceived volume there. In practice, two masters are delivered: one for platforms and another for the booth with a much higher loudness target.

QC: quality control before delivery

Quality control is a checklist, not an impression. Always check the same points: integrated and short-term LUFS within the planned range, true peaks below ceiling, no clipping after MP3 or AAC encoding (listen to the encoded file, not just the WAV), and a mono pass confirming the image survives.

Then the non-measurement checks: exact beginning and end without stray metronome or truncated reverb tail; final fade without cutting the last transient; filename matching the version (date and numbering); and the sent master being the same file heard in final QC, with a hash check if supported. For platform delivery, check title, credits, and ISRC; if also preparing a club version, save it separately under its own name.

Final listening takes place in three conditions: reference monitors, small speaker or consumer headphones, and mono. Only then is the master deliverable.

Next step

For mastering as a whole, the complete level/export framework is in Mastering guide: LUFS, True Peak, limiting, and export, and the measurements underpinning this entire process in dBFS and LUFS meters in gain staging.