The chain, and why the order is not arbitrary

The chain, and why the order is not arbitrary

A voice track goes through the same sequence of stages in almost every professional context. The order is not convention; each stage changes what the next one sees, and getting it wrong makes every subsequent decision harder.

1  gain staging        get the track to a sensible working level
2  editing and repair  cuts, room tone, clicks, plosives, loud breaths
3  broadband noise     the least reduction that solves the problem
4  subtractive EQ      high-pass, remove resonances and mud
5  compression         control the dynamic range
6  de-essing           after compression, because compression causes it
7  additive EQ         presence and air, on a now-stable signal
8  saturation          optional, for density
9  reverb / space      on a send
10 bus processing      gentle glue compression across the mix
11 loudness and limit  the final target and true-peak ceiling

Why each boundary is where it is

Repair before processing. A click, a plosive or a chair creak fed into a compressor triggers gain reduction that ducks the words around it. Fix the event first and the compressor never sees it.

Noise reduction early. Every later stage (EQ boosts, compression, saturation) amplifies noise along with signal. Removing it first means you are not fighting it four times.

Subtractive EQ before compression. A compressor responds to total energy, and low-frequency rumble carries a great deal of it. Compress before high-passing and the rumble drives the gain reduction, so the voice ducks every time a lorry passes. High-pass first and the compressor responds to the voice.

De-essing after compression. Compression turns down the loudest parts; sibilants are rarely the loudest parts, so afterwards they stand relatively higher. De-essing first means re-creating the problem in the next stage.

Additive EQ after compression. Boost 4kHz first and the compressor now triggers harder on the boosted band, which partly undoes the boost. Compress into a stable signal, then shape it.

Reverb last among the voice processing. Reverb should be applied to a finished voice. Compressing after a reverb squashes the tail and pumps the room; EQing after it changes the space’s tone rather than the voice’s.

These are defaults, not laws. A deliberate departure (a compressor before EQ to catch a specific dynamic problem, a touch of reverb before a saturator for a particular effect) is fine as long as you can say why.

Gain staging through the chain

Each stage should hand on a level near where the next expects it. Plugins modelled on analogue equipment often have a sweet spot around −18 dBFS, and a signal arriving 15dB hotter will behave differently from the way its designer intended.

Practically: use a gain trim at the start, and use the output or make-up control on each plugin to leave the level roughly as it arrived. If the level is climbing 6dB at every stage, you will discover it at the limiter and have to unpick everything.

Track, bus, master

Three layers, each with a job.

Track processing solves problems specific to one voice: this speaker’s sibilance, this microphone’s proximity, this room’s resonance.

Bus processing makes several elements behave as one. A dialogue bus with 1 to 2dB of gentle compression across all speakers makes a conversation feel like one scene rather than a series of separate recordings.

Master processing is loudness and safety only: the limiter, the true-peak ceiling, perhaps a very gentle overall EQ. Anything more on the master is a sign that a decision should have been made further up.

The commonest structural mistake is doing bus work on individual tracks: compressing each speaker hard to make them match, rather than balancing them and letting a bus compressor do the last decibel.

Working order, as opposed to signal order

The signal chain and the order you build it in are different.

1. Edit first, entirely. Get the content right before touching a plugin. There is no point in EQing a line you are going to cut.
2. Balance with faders, no processing. Set relative levels. A surprising amount of what people reach for a compressor to fix is a fader problem.
3. Then process, cheapest move first: a fader, then an EQ cut, then a compressor, and only then anything more elaborate.
4. Then automate the things a static process cannot handle.
5. Then measure and set the delivery loudness.

Mixing in this order means each tool is solving what is left rather than what an earlier tool created.

Two habits worth building

Bypass everything periodically. After an hour, you will have adapted to your own decisions. A raw-versus-processed comparison, level-matched, tells you whether you have improved the track or merely changed it, and sometimes you have made it worse and become used to it.

Leave it overnight. Fresh ears catch in thirty seconds what tired ears miss for an hour. Every experienced mixer says this and it remains the least followed advice in audio.

The free path

Ardour and Reaper give you full bus routing, plugin chains and automation. Audacity does effects destructively per clip, which makes it awkward for chain work but perfectly usable for a single voice track. The LSP and Calf plugin suites cover EQ, compression, de-essing, gating and limiting. DaVinci Resolve’s free Fairlight page gives a complete mixing environment with a fixed channel strip in the professional order, which is itself a useful thing to look at, because the order it enforces is the one in this lesson.

The limitation

A chain is a starting structure, not a recipe. Every recording arrives with its own problems, and a template applied blindly does harm: a high-pass at 100Hz on a deep voice recorded at a distance removes body it needed; a compressor on a track already even in level only adds noise and pumping.

The discipline that prevents this is to ask, at each stage, what problem this stage is solving on this track. If the answer is “it is what I always do”, switch it off and see whether anybody notices.