what is compression
what audio compression actually does, how it controls dynamic range, and why every modern mix depends on it. a practical guide for music producers.
the vocal that disappears
you are mixing a vocal. during the quiet verse, the singer is barely audible behind the guitars. during the chorus, the same vocal jumps 12 dB louder and clips the bus. you reach for the fader, but the problem is not overall level. the problem is dynamic range: the difference between the quietest and loudest moments is too wide for the vocal to sit consistently in the mix.
this is the problem audio compression solves. before compressors existed, engineers literally moved faders with their hands, riding the volume in real time. compression automates that.
compression reduces the dynamic range of a signal. it makes loud parts quieter, or quiet parts louder, or both, depending on the type. the result is a signal that stays more consistent in level, sitting in the mix without disappearing in the quiet moments or jumping out in the loud ones.
every modern mix uses compression. on individual tracks, on buses, on the master. it is as fundamental to mixing as EQ. but where EQ changes what frequencies are present, compression changes how they behave over time.
key takeaway
compression does not change what frequencies are in your signal. it changes how loud those frequencies are allowed to be, moment by moment. the frequency content stays the same. the dynamic behavior changes.
dynamic range: the core concept
dynamic range is the difference in decibels between the quietest and loudest parts of a signal. a whispered verse might sit at -30 dBFS. a belted chorus might hit -6 dBFS. that is a 24 dB dynamic range.
in a live performance, 24 dB of dynamic range is natural and exciting. in a recorded mix competing with other tracks on a streaming platform, 24 dB of dynamic range means the quiet parts are inaudible in a car and the loud parts are painful on headphones. the listening environment cannot accommodate the full dynamic range of the performance.
compression narrows that range. a compressor set to reduce peaks by 6 dB turns a 24 dB dynamic range into an 18 dB dynamic range. after applying makeup gain to bring the overall level back up, the quiet parts are now 6 dB louder relative to the loud parts. the verse is audible. the chorus is controlled.
this is not about making things louder. it is about making the relationship between quiet and loud more manageable for the listening context.
decibels are logarithmic
a 6 dB reduction is roughly half the amplitude. a 12 dB reduction is a quarter. your ears perceive loudness logarithmically, which is why audio uses decibels instead of linear amplitude values. a 3 dB change is the smallest difference most listeners notice. a 10 dB change sounds roughly “twice as loud.” these perceptual realities are why compressor controls are calibrated in dB.[^1]
downward compression
the most common type. downward compression reduces the level of signals that exceed a threshold.
set a threshold at -18 dBFS. anything louder than -18 dBFS gets turned down. anything quieter passes through unchanged. the amount it gets turned down depends on the ratio: at 4:1, a signal that exceeds the threshold by 8 dB is reduced to only 2 dB above the threshold.
the result: loud peaks are pulled closer to the average level. the overall dynamic range shrinks from the top down. this is what most people mean when they say “compression.”
downward compression is the standard approach for controlling vocal dynamics, taming drum transients, evening out bass guitar levels, and gluing a mix bus together. it is the foundation.
upward compression
upward compression does the opposite. instead of turning loud signals down, it turns quiet signals up.
set a threshold at -24 dBFS. anything quieter than -24 dBFS gets boosted toward the threshold. anything louder passes through unchanged. the ratio determines how aggressively the quiet content is lifted.
the result: quiet details that were buried become audible. room ambience, cymbal sustain tails, vocal breath, the decay of a reverb, the subtle string noise on a guitar. upward compression recovers content that was always in the recording but too quiet to hear in context.
this is a fundamentally different tool from downward compression. downward compression controls peaks. upward compression reveals detail. they are complementary, not interchangeable.
upward compression is less common in traditional mixing but increasingly important in modern production. per-band upward compression, where the boosting happens independently at each frequency, can bring up high-frequency air without also bringing up low-frequency rumble. this frequency-selective approach is what separates tools like spectral compressors from simple broadband upward compression.
key takeaway
downward compression reduces peaks from above the threshold. upward compression lifts detail from below the threshold. most compressors do downward. the combination of both, especially when applied per frequency band, gives you dynamic control that neither approach achieves alone.
why compression sounds musical
compression does more than level-control. the way a compressor responds to transients, sustain, and dynamics creates a sonic character that shapes how music feels.
transient shaping
when a drum hit arrives, the initial spike (the transient) can be 20 dB louder than the sustain that follows. a fast compressor catches the transient and reduces it. a slow compressor lets the transient through and only compresses the sustain. the difference changes whether drums sound punchy and immediate or smooth and sustained.
this is not a side effect. engineers choose compressor timing specifically to shape transients. letting the attack through while compressing the sustain is how you get a drum sound that punches through a dense mix. compressing the attack is how you push drums further back.
perceived loudness
your ears judge loudness based on the average level, not the peak level. when a compressor reduces peaks by 6 dB and you add 6 dB of makeup gain, the peaks are back where they started but the average level is 6 dB higher. the signal sounds louder even though the peak level has not changed.
this is why compression is central to the loudness wars and why every mastering chain includes it. controlled dynamic range allows higher average levels, which translates to perceived loudness on streaming platforms, in headphones, and in car speakers.[^2]
sustain and body
compression increases the apparent sustain of a signal. by reducing the loud attack and boosting the overall level, the quieter sustain tail becomes more prominent. a short, snappy bass note becomes a round, sustained one. a guitar chord that decays quickly rings out longer. the notes are not actually longer. they are just louder relative to the attack.
tip
listen to compression with your eyes closed. the gain reduction meter tells you what is happening technically, but your ears tell you what is happening musically. a compressor working 6 dB of gain reduction might sound like it is doing nothing if the timing is right, or it might sound like it is destroying the signal if the timing is wrong. trust the sound, not the numbers.
the signal flow of a compressor
every compressor, from a vintage 1960s hardware unit to a modern plugin, follows the same basic signal flow.
the detector measures the signal level. this can be the peak level (the absolute maximum at any instant), the RMS level (the average power over a short window), or something more complex like crest factor (the ratio of peak to RMS). the detection method determines how the compressor responds to transients versus sustained content.
the gain computer compares the detected level to the threshold and calculates how much gain reduction to apply. this is where the ratio, knee, and threshold settings interact to determine the compression curve.
the gain stage (called a VCA in hardware, a multiplier in software) applies the calculated gain reduction to the audio signal. this is the part that actually changes the level.
makeup gain adds a fixed amount of gain after compression to bring the overall level back up. some compressors do this automatically (auto-makeup gain). others leave it to you.
the critical insight: the detector and gain computer operate on a control signal (the envelope). the audio signal passes through the gain stage unmodified except for the level change. this separation between the detection path and the audio path is what allows compression to be transparent. the audio signal is never “processed” in the way an EQ or saturator processes it. it is simply made louder or quieter.[^3]
feed-forward vs feedback
in a feed-forward compressor, the detector measures the input signal before the gain stage. in a feedback compressor, the detector measures the output signal after the gain stage. feed-forward designs give more predictable, precise compression. feedback designs create a self-regulating loop that tends to sound smoother and more musical. most modern compressors are feed-forward. the classic 1176 and LA-2A are feedback designs.
broadband, multiband, and spectral
not all compressors act on the full signal at once. the three main approaches each control dynamics at a different level of frequency detail.
broadband compression applies one gain change to the entire signal. when the overall level exceeds the threshold, everything gets turned down equally: bass, mids, highs, all at once. this is the simplest and most transparent approach when the dynamic range problem is uniform across the spectrum.
multiband compression splits the signal into 3 to 6 frequency bands using crossover filters. each band has its own compressor with its own threshold, ratio, and timing. a loud bass note triggers compression in the low band without affecting the highs. this solves the pumping problem (cymbals ducking every time the kick hits) but introduces crossover artifacts at the band boundaries.
spectral compression decomposes the signal into hundreds of frequency regions using FFT analysis. each region has its own independent compressor. a resonance at 800 Hz triggers compression at 800 Hz. the vocal at 2 kHz, the cymbal at 12 kHz, and the bass at 80 Hz are processed independently, with no crossover artifacts.
the trade-off: more frequency resolution means more transparent compression on complex material, but also more latency and CPU. broadband compression runs in real time with essentially no latency. spectral compression typically adds 50 to 100 ms of latency due to the FFT analysis.
where compression fits in your mix
compression serves different purposes at different points in your signal chain.
individual tracks
this is where you solve specific problems. a vocal with too much dynamic range. a bass guitar that disappears on quiet notes. a snare that is inconsistent between hits. the goal is to make each track sit more consistently so it plays well with the rest of the mix.
group buses
drum bus compression is the classic example. compressing the mixed drum kit together makes the individual elements (kick, snare, toms, overheads) interact as a single instrument. the shared gain reduction creates a sense of cohesion that individual track compression cannot achieve.
mix bus
gentle compression on the entire mix ties the elements together. 1 to 3 dB of gain reduction is typical. the goal is glue, not control. too much mix bus compression flattens the dynamics of the entire song. the right amount makes the mix feel like a single cohesive piece of music rather than a collection of separate tracks.
mastering
mastering compression is the final dynamic shaping before your music reaches listeners. it is gentle, transparent, and focused on consistency across the entire track. you typically use broad, slow compression here to control the overall dynamic envelope without altering the mix balance.
heads up
compression at every stage compounds. 3 dB of compression on a vocal, plus 2 dB on the vocal bus, plus 2 dB on the mix bus, plus 1 dB in mastering is 8 dB of total compression. each stage should be aware of what came before and what comes after. over-compressing at any one stage makes it harder for every subsequent stage to sound natural.
compression is not a volume knob
the most common mistake is treating compression like a level control. turning up the ratio until the meters look right, then adding makeup gain until it sounds loud.
compression is a dynamics shaping tool. it changes how your audio breathes, how transients hit, how sustain decays, and how the quiet details relate to the loud ones. the numbers on the gain reduction meter matter less than what you hear.
start every compression decision with a question: what is the dynamic problem i am trying to solve? if the answer is “the vocal keeps disappearing,” you need a moderate ratio and threshold that catches the quiet dips. if the answer is “the snare transients are too sharp,” you need a fast attack and moderate ratio. if the answer is “the mix bus feels disconnected,” you need gentle ratio and slow timing.
the right compressor settings come from understanding the problem, not from memorizing preset values.
frequently asked questions
frequently asked questions
what does compression do to audio?
compression reduces the difference between the loudest and quietest parts of your audio. it turns down loud peaks (downward compression) or turns up quiet parts (upward compression), making the overall signal more consistent. the result is a sound that sits more steadily in a mix without wild volume swings.
is compression the same as limiting?
a limiter is a compressor with a very high ratio, typically 10:1 or higher. both reduce dynamic range, but a compressor shapes dynamics gradually while a limiter acts as a hard ceiling. you use compression for tonal shaping and consistency, and limiting for peak control and loudness maximization.
do you need compression on every track?
no. compression is a tool, not a rule. some tracks benefit from it (vocals, drums, bass), while others sound better left alone (ambient pads, room mics, anything where dynamics are the point). use compression when the dynamic range of a track causes problems in the mix, not as a default insert.
what is the difference between downward and upward compression?
downward compression reduces the level of signals above a threshold, making loud parts quieter. upward compression boosts signals below a threshold, making quiet parts louder. downward compression controls peaks. upward compression recovers detail. most compressors are downward, but per-band upward compression is increasingly used for recovering ambience and high-frequency air.
can compression make a mix sound worse?
yes. over-compression removes the natural dynamics that give music its energy and emotion. common symptoms include pumping (audible volume ducking), loss of transient impact, and a flat, lifeless sound. the goal is to control dynamics enough that the mix is consistent, without removing the dynamics that make it musical.
references
a note from the developer
this guide exists because i spent three years building a compressor and realized the foundations matter more than the features. when i started designing KERN PUSH, i thought the hard part would be the spectral processing, the ERB bands, the per-bin gain smoothing. those were hard. but the hardest part was understanding what compression actually needs to do for a mix, not what it technically can do.
i am a solo developer in copenhagen, and these guides are my attempt to share what i have learned. not just the theory, but the decisions you face when you are actually mixing music and reaching for a compressor.
if you have a compression workflow that contradicts anything here, i want to hear about it. jonas@kernaudio.io. i learn from every conversation.
try it yourself
KERN PUSH: three compression characters across 40 spectral bands. $29, no iLok, no subscription.