Attack, Decay, Sustain, Release

Two sounds can contain exactly the same frequencies, at exactly the same levels, and be completely unrecognisable as relatives of one another. Reverse a piano note and it stops sounding like a piano — nothing about its harmonic content changed, only the order in which it arrived.

What changed is the envelope: the shape of the sound's loudness over time. It is the second half of what makes a sound identifiable, and it is the thing every compressor, gate, limiter and sampler you will ever use is actually operating on.

Two different timescales

It helps to separate two things that both describe amplitude but on wildly different scales.

The waveform is the microscopic picture: individual pressure swings happening hundreds or thousands of times per second. Zoom all the way into a DAW and you eventually see it — actual cycles.

The envelope is the macroscopic picture: the overall outline of how loud the sound is across its whole duration, from the moment it starts to the moment it dies away. Zoom out in a DAW and the individual cycles blur into a solid block — and the shape of that block is the envelope.

Both are amplitude. They just describe it at different zoom levels, and almost everything in dynamics processing happens at the envelope level. When you set a compressor's attack to 10 milliseconds, you are telling it to respond to the outline rather than to individual cycles. When you set it too fast, it starts chasing the cycles themselves, and the result is distortion — a problem the compression lessons return to in detail.

The four stages

The standard model breaks an envelope into four parts. It comes from synthesiser design, where it maps exactly onto four controls, but it is a useful description of natural sounds too.

• Attack — from silence to the loudest point. How quickly the sound arrives.
• Decay — from that initial peak down to whatever level it settles at.
• Sustain — the level it holds while energy is still being supplied. Note that this one is a level, not a time, which is the detail everyone gets wrong first.
• Release — from the moment the energy stops to silence.

One honest caveat: real sounds do not obey this neatly. A struck cymbal has no meaningful sustain stage at all — it peaks and decays for twenty seconds. A bowed note may never decay. ADSR is a useful abstraction that happens to fit synthesiser hardware perfectly and natural sound approximately. Use it as vocabulary, not as law.

Attack is where the identity is

The attack is the most information-dense part of almost any sound. It is where the physical mechanism of the instrument is most exposed — the stick hitting, the reed catching, the string being released by the plectrum, the hammer noise before the piano string speaks.

Fast attacks reach full level almost instantly:

• a rimshot
• a hi-hat foot chick
• a bottle cap cracking open
• a hand clap in a dry room
• a hard consonant at the start of a sung word

Slow attacks take a noticeable amount of time to build:

• a bowed double bass
• a synth pad swelling in
• a cymbal struck with a soft mallet rather than a stick
• a kettle coming to the boil
• a harmonium note as the bellows take up

The practical consequence appears immediately in mixing. Fast-attack sounds cut through a dense arrangement because your ear locks onto sudden changes. Slow-attack sounds sit underneath without competing, which is exactly why pads and swells work as bed layers and why a bowed part can carry a lot of level without ever pulling focus.

It also means a fast-attack sound is fragile in a way a slow one is not. Put a compressor with a 1 ms attack across a rimshot and you remove the very spike that identified it, leaving a dull thud with the same frequency content and none of the character. Put the same compressor across a pad and almost nothing happens, because there is no transient to catch.

Sustain is a question about energy

A sound sustains for as long as something keeps feeding it. A bowed string sustains because the bow keeps supplying energy. An organ note sustains because air keeps flowing. A struck piano string does not really sustain — it decays slowly, because all the energy it will ever have was delivered in one moment by the hammer.

This distinction is why some instruments are easy to mix and others are not. Continuously-driven sounds hold a steady level and stay where you put them. Struck and plucked sounds are loudest at the start and quietest at the end, so a single fader position is always wrong for at least part of the note — which is the actual reason those sources need compression more than sustained ones do.

Release is the room telling on itself

The release is the most overlooked stage and the most useful one for a recording engineer, because it contains information about the space rather than the source.

When a sound source stops, what you go on hearing is the room emptying out — energy still bouncing between surfaces, arriving later and later and quieter and quieter. So the release tells you where the recording happened, before anyone tells you:

• Short release, almost no tail — a small, absorbent space. A treated booth, a bedroom full of soft furnishings, a car.
• Long release with a smooth decay — a large reflective space. A hall, a stairwell, a temple.
• Long release with distinct repeats — a large space with strong parallel surfaces. You are hearing individual echoes rather than a diffuse tail.
• Essentially no tail at all — outdoors, where reflections have nothing to come back from.

This is a skill worth building deliberately, and the exercise at the end of this lesson is designed for it: record the same clap in three spaces and you will find you can identify rooms by their release for the rest of your career. It is also the fastest way to hear whether an added reverb is fighting the room already on the recording — if the recorded release and the artificial one disagree about the size of the space, the ear notices immediately even if it cannot say why.

Two pieces of working vocabulary come from here: the beginning of a sound is often called its head and the end its tail. You will hear both constantly in editing and mastering.

A transient is a broadband event

Something that is easy to miss: a sharp attack is not just a loudness event, it is a frequency event.

A sudden change in amplitude requires energy across a wide range of frequencies to produce it. The faster the change, the wider the range needed. A perfectly instantaneous click contains, in principle, every frequency at once — which is why a click is such a useful test signal, and why a badly placed edit produces a click that you hear as a broadband tick rather than as a note.

Two consequences follow, and both come up constantly.

First, anything that removes high frequencies softens attacks. A low-pass filter, a dull microphone, a lossy codec at a low bitrate, or simply distance through air — all of them blunt transients as a side effect of removing top end, and the result reads as "further away" or "softer" rather than as "less treble".

Second, anything that blunts attacks costs you high-frequency content. A fast compressor attack does not just reduce the peak; it removes some of the broadband energy that made the attack sharp. This is why heavily compressed drums often need top-end EQ afterwards — the compressor did not touch an EQ band, but it removed the transient that was producing the perception of brightness.

Envelope and how loud something seems

Two sounds can peak at exactly the same level and be nothing like equally loud, and the envelope is why.

Your ear does not respond instantly. It integrates over roughly a couple of hundred milliseconds when judging loudness, so a very brief peak barely registers as volume even when a meter shows it hitting the ceiling, while a sustained sound at a lower peak level can seem considerably louder.

This is the whole reason peak metering and loudness metering disagree, and the reason a heavily limited master sounds louder without its peaks being any higher — the limiting raised the average, which is what you hear, while leaving the peak where it was. It is also why a sound with a sharp attack cuts through an arrangement at a lower fader setting than a sustained one needs: it is not competing for the same perceptual space.

Tools that work on the envelope directly

Beyond compressors and gates, a couple of tools operate on envelope shape explicitly, and they are worth knowing about early because they solve problems EQ cannot touch.

A transient designer gives you separate control over attack and sustain, independent of threshold. Turn the attack up and a drum hits harder; turn the sustain down and the room around it shrinks without a gate's abruptness. Because it is responding to the shape rather than to an absolute level, it works consistently across a performance whose level moves around.

Envelope followers and envelope filters take the shape of one signal and use it to control something else entirely — a filter cutoff, another track's level, a send amount. Once you can see envelopes, using one as a control source stops being an exotic idea.

This is the compression lesson in disguise

Everything above is why the dynamics lessons later in the curriculum will feel like a continuation rather than a new subject.

A compressor has two timing controls, and both of them are envelope controls:

• Attack decides how much of the sound's attack survives. Set it slow and the transient passes through before the compressor reacts, so the hit stays sharp and only the body gets controlled — which is precisely how you make drums sound bigger rather than smaller. Set it fast and you catch the transient itself, flattening the hit.
• Release decides what happens to the tail. Set it too fast and the compressor recovers audibly between events, so you hear the room level lurching up and down. Set it too slow and it never recovers before the next hit, so everything stays clamped.

A gate is the same idea from the other direction: it decides how quickly to open at the start of a sound and how quickly to close on the tail — and the reason a badly set gate sounds so obviously wrong is that it truncates a release the ear expected to hear complete.

Once you can look at a waveform and read its envelope, dynamics settings stop being trial and error. You are shaping a shape you can already see.

And the sampling lesson too

The same knowledge is what makes sample editing possible. Where you place a sample's start point determines whether the attack survives. Whether you loop the sustain or let it decay naturally determines whether the sound can be held indefinitely. Whether you truncate the tail determines whether the room comes with it.

It is also why a single sample transposed across a keyboard sounds wrong in a specific way: transposing changes the playback rate, so the whole envelope stretches or compresses along with the pitch. A note played an octave down does not just sound lower, it takes twice as long to happen — and a piano's attack does not actually get slower when you play a lower note. The sampling lessons deal with the fixes.

A shape you can now hear

Spend a day listening for envelopes rather than for pitches and something shifts. You start noticing that a room's release tells you its size before you have looked around it. That a badly compressed snare has lost its head, not its level. That the reason a sampled string section sounds fake is almost always a wrong attack rather than a wrong tone.

Frequency content tells you what a sound is. The envelope tells you how it happened — and your ear, which spent a very long evolutionary period working out whether a sudden noise mattered, pays close attention to how things happen.

What to practice

• Record five sounds with deliberately different envelopes. Look at each waveform and predict, from the shape alone, which will survive a fast-attack compressor intact. Then test it.
• Record one hand clap in a bathroom, a bedroom and outdoors. Do not change anything else. The only thing that differs is the release, and it will tell you the room instantly.
• Take a sustained note and reverse it. Describe what changed — it is only the envelope, since the harmonic content is identical.
• Find a sound in your library whose attack you cannot identify with the first 30 ms removed. That is how much identity lives in the attack.