How to Get Audio Compression and Dynamics Right

Audio compression and dynamics processing is the craft of controlling the distance between the loudest and quietest parts of a signal. The tools are familiar to anyone who has opened a digital audio workstation: compressors that turn down what rises above a set level, limiters that stop peaks outright, gates that close off the quiet gaps, and expanders that push low-level material further down. Used well, they make a voice sit steadily over music, keep a podcast comfortable in a car, and let a mix translate from studio monitors to a phone speaker. Used badly, they flatten performances, raise the noise floor and create the audible "breathing" that listeners notice even when they cannot name it.

This matters to more people than mix engineers. Marketing teams approving a brand video, producers assembling a weekly podcast, educators recording lectures and in-house teams cutting social clips all make dynamics decisions, whether they touch a compressor themselves or sign off on someone else's work. Getting audio compression and dynamics right is the difference between content that sounds confident and content that sounds either timid or squashed.

Dynamics processing is also the most misapplied tool in audio, for a simple reason: it is easy to make something louder and hard to notice what was lost in the process. Louder almost always sounds "better" on first listen, so a heavy-handed setting wins a casual comparison and then wears the listener out over ten minutes. This guide is organized as a checklist. The master list comes first; each section after it explains why an item matters and how to do it well, with starting settings, a worked example and the trade-offs that decide between approaches.

Name the Reason Before You Reach for a Compressor

The single most useful habit in dynamics work is to say, in one sentence, what a processor is for before you insert it. "Keep the host's level steady when she leans back from the microphone." "Catch the snare hits that jump 6 dB above the rest of the kit." "Stop the final master from exceeding the true-peak ceiling." Each of those sentences implies a different threshold, ratio, attack and release. A compressor inserted "because vocals always get compression" or because a channel-strip preset loaded one has no target, and a processor without a target tends to be set until it sounds different, which usually means louder and denser.

Naming the reason also tells you when to stop. If the job is to even out a speaker who varies by 10 dB between sentences, you are done when the variation is small enough that the listener no longer reaches for the volume control. You are not done when the meter shows a certain amount of gain reduction or when the track "sounds compressed." The goal sits in the listener's experience, not on the plugin interface.

Common legitimate reasons

• Level consistency: reducing the difference between loud and soft phrases so speech or a lead instrument stays intelligible over a bed of music or background noise.
• Peak control: catching short, occasional spikes so the rest of the signal can be raised without overloading the next stage.
• Tone and envelope shaping: using attack and release deliberately to emphasize or soften the front of a note, such as adding punch to a kick drum or smoothing a plucked guitar.
• Glue: gentle bus compression that makes several sources respond together, so a group of drums or a dialogue-and-music mix feels like one piece.
• Delivery compliance: keeping a finished program within a loudness range and peak ceiling for a broadcaster, platform or venue.
• Noise management: gating or expanding to reduce room tone, headphone bleed or air-conditioning hum between phrases.

If a processor does not map to one of these, or to a similarly specific reason you can articulate, bypass it and listen. Very often nothing gets worse, and the mix gets a little more open.

Clean Up the Signal Before Dynamics Processing Touches It

Compressors react to whatever arrives at their input. If that input contains hum, a clipped consonant, a rumble from a desk being bumped or a level that is 20 dB too low, the compressor will faithfully make those problems more prominent. Compression raises the quiet parts relative to the loud parts; the noise floor lives in the quiet parts. A recording with audible room tone at a moderate level can end up with intrusive room tone after 8 dB of gain reduction and matching make-up gain.

So the order of operations matters. A sensible chain for spoken word typically runs: clip repair and de-clicking, noise reduction, high-pass filtering and corrective EQ, then dynamics, then any tonal enhancement, then limiting at the very end. Music chains vary more, but the principle holds: fix defects first, shape dynamics second, and catch peaks last.

Upstream checks worth doing first

• Gain staging: record and import at sensible levels, with peaks comfortably below full scale. Interface and cabling problems, including hum picked up on long unbalanced runs, are covered in our guide to balanced and unbalanced audio connections.
• Clipping: a compressor cannot un-clip audio. Squared-off waveforms need dedicated repair before any dynamics stage; see how to approach repairing clipped audio for the process.
• Low-frequency energy: rumble, plosives and handling noise carry a lot of energy that the compressor's detector will react to, causing gain reduction that has nothing to do with the voice. A high-pass filter before the compressor, or a side-chain filter inside it, prevents that.
• Clip gain: for spoken word, manually lowering a handful of very loud words or raising a few very quiet phrases with clip gain or volume automation often does more than any compressor, and it has no attack or release artifacts at all.

Set Threshold and Ratio by Watching Gain Reduction

The threshold is the level above which the processor acts. The ratio is how much the signal above that threshold is reduced. With a 4:1 ratio, a signal that rises 8 dB above the threshold comes out only 2 dB above it, so the compressor has applied 6 dB of gain reduction. With 2:1, the same 8 dB overshoot comes out 4 dB above threshold, for 4 dB of gain reduction. That arithmetic is simple, but in practice the settings on the knobs are a poor guide to what is actually happening, because the amount of processing depends on the input level, which changes constantly.

That is why the gain reduction meter is the most important display on any compressor. It shows the actual amount being applied, moment by moment. A threshold of -20 dBFS means nothing on its own: on a quiet recording it might produce no gain reduction at all, and on a hot one it might produce 12 dB. Always set the threshold while watching that meter and listening, and ask whether the amount and the pattern of gain reduction match the reason you wrote down.

A practical way to set them

1. Start with a moderate ratio around 2:1 to 3:1 for general level control, so threshold changes are easy to hear without extreme effects.
2. Lower the threshold slowly while playing a representative passage, until the meter shows gain reduction only on the louder phrases or hits you intended to control.
3. Read the pattern, not just the peak. Gain reduction should come and go with the material. If the meter never returns near zero, the threshold is too low and you are compressing everything, including the quiet parts you wanted left alone.
4. Adjust ratio to taste. Raise it if loud moments still jump out; lower it if the sound becomes constrained. The threshold and ratio interact, so revisit the threshold after any ratio change.
5. Add make-up gain last, and only enough to match the bypassed loudness for comparison. More on that in the matching section.

Knee, detection and side-chain filtering

Beyond threshold and ratio, three controls change how the compressor behaves. A soft knee begins compressing gradually as the signal approaches the threshold, which sounds smoother and suits vocals and buses; a hard knee applies the full ratio the moment the threshold is crossed, which suits peak control and deliberate punch. Peak detection responds to instantaneous levels and catches fast transients; RMS or average detection responds to sustained energy and behaves more like the ear, which often suits level-riding on voices. A side-chain high-pass filter stops bass energy from triggering gain reduction, which is how you keep a mix bus compressor from pumping every time the kick drum hits. Manufacturer documentation, such as the dynamics references published by iZotope, explains how each of these behaves in a given tool, and it is worth reading for any processor you use daily, because the same label can mean slightly different things across products.

Choose Ratios That Fit the Job, from Gentle Control to Limiting

Ratio is where the family of dynamics processors separates. Low ratios, roughly 1.5:1 to 3:1, give gentle, transparent control and are the default for glue and for most voice work. Middle ratios, roughly 4:1 to 8:1, give firm control for sources with wide swings, such as an energetic vocal or a bass guitar played with a pick. High ratios, from about 10:1 upward, behave increasingly like a limiter. Limiting is compression at a very high ratio, used to stop peaks. It should act briefly and rarely on most material.

The table below collects common starting ranges. They are starting points for your ears, not settings to copy, and every value will move depending on the performance, the microphone, the arrangement and the reason you named.

Notice how modest the gain reduction figures are. A bus compressor doing 1 to 2 dB can make a noticeable difference to how a mix holds together. When you find yourself needing 12 dB or more on one processor, that is a signal to revisit the recording, the clip gain or the arrangement, or to split the work across stages.

Time Attack and Release to the Material's Own Rhythm

Attack and release determine how quickly the processor engages and lets go, and they have more effect on the character of the result than threshold or ratio. They should be set by ear against the rhythm of the material itself: the pace of a speaker's syllables, the tempo of a song, the decay of a drum, the length of a sustained note.

What attack does

Attack is how long the compressor takes to reach its full gain reduction once the signal crosses the threshold. The first few milliseconds of most sounds, the transient, carry a disproportionate share of the information the ear uses to identify and locate them: the consonants that make words intelligible, the stick hitting a drumhead, the pick on a string. A fast attack clamps down on that transient, which makes the sound smoother and further back but also duller and less distinct. A slower attack lets the transient through before gain reduction starts, which preserves or even exaggerates punch and clarity.

What release does

Release is how long the compressor takes to return to no gain reduction after the signal falls back below the threshold. Too fast, and the gain rises and falls within individual waveform cycles on low-frequency material, causing distortion, or it rises audibly between syllables, making the noise floor pump. Too slow, and the compressor is still turning things down when the next quiet phrase arrives, so soft words that follow loud ones get swallowed. A well-set release lets the gain recover in time with the natural gaps in the material, which is why tempo-synced or program-dependent release modes are popular.

Setting them by ear

1. Temporarily exaggerate the effect: lower the threshold and raise the ratio until the compression is obvious. Artifacts are easier to hear when they are large.
2. Sweep the attack from fastest to slowest. Listen to the front of words or hits. Stop where the transients sound natural and clear rather than blunted or spiky.
3. Sweep the release. Watch the gain reduction meter bounce in time with the material. Aim for the needle to return close to zero in the gaps between phrases or beats, without audible swelling of background noise.
4. Return threshold and ratio to their intended, gentler values and listen again in context.

Protect the Transients That Carry Clarity

This item deserves its own place on the checklist because fast attack settings that remove the transients that carry clarity are among the most common and least noticed mistakes. The result rarely sounds broken. It sounds slightly muffled, slightly distant, as though someone had added a gentle high-cut filter. The natural response is to reach for an EQ and boost the top end, which adds harshness and noise to fix a problem the compressor created.

Transients also carry the sense of energy in music. A drum mix squeezed with fast attack on every channel and again on the bus loses its impact; turning it up only makes it louder and flatter. The same holds for acoustic guitar, piano and percussion. In spoken content, a voice with softened consonants forces the listener to work harder, which is a real problem for e-learning, audio guides and any content played in noisy spaces.

Ways to control peaks without losing transients

• Slower attack plus a separate fast peak catcher: let the main compressor use a slower attack, and put a fast, high-ratio stage afterward that only touches the rare extreme peaks.
• Parallel compression: blend a heavily compressed copy underneath the untouched signal. The dry path keeps the transients; the compressed path lifts the body and sustain. Many compressors include a mix or dry/wet control that does this in one plugin.
• Clip gain or automation for the outliers: if only a few moments are too loud, turning those moments down by hand costs nothing in transient detail.
• Transient-aware tools: dedicated transient shapers adjust attack and sustain directly, independent of level, and can restore some punch where compression has softened it.

This is also why content made for noisy playback environments, such as retail floors, events and gyms, needs particular care. Density helps in those spaces, but intelligibility depends on transients. Our article on audio for trade show booths covers how playback conditions change those decisions.

Split Heavy Work Across Several Gentle Stages

Using several gentle stages rather than one aggressive one is established good practice, and it is the key trade-off in most dynamics chains. Asking one compressor to deliver 12 dB of control forces it into settings where its artifacts become audible: pumping, dulled transients and a sense that the sound is pressed against a ceiling. Two or three stages each doing 3 to 4 dB, each optimized for a different part of the job, usually sound more transparent while delivering the same overall control.

A typical serial approach on a vocal or narration might be a first compressor with a slower, program-dependent character that rides the overall level at a low ratio; a second, faster stage at a higher threshold that catches the remaining peaks; and, at the end of the full mix, a limiter that touches only the loudest moments. On a music master, the equivalent is gentle bus compression during the mix, then light compression and limiting in mastering, rather than one heavy limiter doing everything.

The practical rule is to add a stage only when a single stage is audibly struggling. If one compressor at 3 dB of gain reduction does the job, you do not need a second. If one compressor at 10 dB sounds strained, splitting the work is almost always the better answer than a different preset. Document what each stage does in your session notes; our guide to archiving audio sessions explains how to keep that information recoverable when a project is reopened months later.

Manage the Quiet End with Gates and Expanders

Compressors and limiters work on the loud end of the signal. Gates and expanders work on the quiet end, and they are as much a part of audio compression and dynamics as the tools that turn things down. An expander increases the dynamic range below its threshold: when the signal falls under the threshold, it is turned down further, by an amount set by its ratio. A gate is an expander taken to the extreme, closing almost completely when the signal drops below the threshold.

Their main uses are practical. On a multi-microphone recording such as a panel discussion or a meeting, each open microphone adds room tone and picks up the other speakers slightly late, which smears clarity. Gently expanding each microphone when its speaker is silent reduces that build-up. On drums, gates reduce bleed between close microphones. On voice recorded in a less-than-ideal room, a downward expander can lower background noise between phrases, which also counters the noise-floor lift that compression causes.

Settings that avoid choppy results

• Prefer range over full closure. Set a range or floor of perhaps 6 to 15 dB rather than muting completely. Listeners hear a noise floor that disappears and reappears far more readily than one that is merely lower.
• Use hysteresis if available. A gap between the opening and closing thresholds prevents chattering when a signal hovers near the threshold.
• Set hold and release for word endings. A release that is too fast clips the tails of words and the decay of notes; a short hold time keeps the gate open through natural micro-pauses.
• Use look-ahead where available, so the gate opens slightly before the first syllable instead of chopping its start.

For recurring formats with many microphones, such as meeting room audio or lecture capture, automatic mixers and gain-sharing systems do a version of this in real time. In post-production, you have the advantage of being able to set it precisely for each voice and check it against the whole program.

Keep the Limiter as a Safety Net, Not a Loudness Machine

A limiter at the end of the chain has one honest job: to stop peaks from exceeding a ceiling, typically the true-peak maximum in a delivery specification. When it touches only the occasional loudest moment by a decibel or two, it is close to inaudible. The problem begins when the limiter is used as the main tool for making a program louder: input gain pushed up until a loudness meter reaches a target, with the limiter doing 6, 8 or 10 dB of gain reduction on every beat or syllable.

Limiting hard to reach a loudness target produces a recognizable set of symptoms. Transients are flattened. Low-frequency energy distorts, because the limiter is working within individual waveform cycles. The mix loses depth, as quiet reverb tails and background details are pulled up to meet the foreground. Intersample peaks can still exceed the ceiling on playback after lossy encoding, if the limiter is not true-peak aware. And on platforms that normalize loudness, the extra effort is largely wasted: the heavily limited version is turned down to the same playback level as a more dynamic one, and then simply sounds smaller.

The fix is to return to the earlier stages: gentle leveling on individual sources, bus compression, clip gain on outliers, and EQ that removes unnecessary sub-bass. With those in place, a limiter doing 1 to 3 dB is usually all that is needed to hit a sensible target.

Compare Processed and Bypassed Audio at Matched Loudness

This is the rule that keeps every other item on the checklist honest: compare at matched loudness, or the louder version always wins. Human hearing is more sensitive to midrange and high frequencies as level rises, so a version that is even 1 dB louder tends to sound fuller, brighter and more detailed. Because most compressors add make-up gain, and because many apply it automatically, engaging a compressor usually makes the signal louder, and the comparison is rigged in its favor before you have judged anything else.

Judging compression without matching levels is one of the named mistakes for good reason. It is how chains accumulate processors that do nothing useful, how mixes become fatiguing and how masters end up over-limited. Mixing and processing guides from DAW makers, such as those from Steinberg, routinely make the same point about comparing processing at equal level.

How to match levels reliably

1. Engage the compressor and set it as you intend to use it.
2. Play a representative section of at least 20 to 30 seconds, not a single loud moment.
3. Measure the integrated or short-term loudness of the processed and bypassed versions with a loudness meter, and adjust make-up or output gain until they match within a fraction of a decibel.
4. Use the plugin's auto-gain or level-matched bypass if it has one, but verify it with a meter; automatic compensation is an estimate.
5. Switch between the two repeatedly, ideally without looking at which is which. Ask a specific question: are words clearer, is the level steadier, did the drums lose impact?

Deliver to the Loudness Specification, with a Worked Example

The last item on the checklist brings everything together. Most finished audio now has to meet a loudness specification: a target integrated loudness measured in LUFS (or LKFS in the North American broadcast standard), a maximum true-peak level, and sometimes a loudness range. Broadcast standards such as EBU R128 in Europe, with a target of -23 LUFS, and ATSC A/85 in the United States, with a target of -24 LKFS, are strict and well documented. Streaming, podcast and social platforms publish their own reference levels, which change over time, so check the current figure for each destination rather than relying on a number you remember. Technical bodies including the Audio Engineering Society publish recommendations on loudness and dynamic range for different delivery contexts that are worth consulting when a specification is unclear.

Meeting a specification without over-compression is mostly a matter of arriving at the final stage with dynamics already under control, so the limiter has little to do.

Worked example: a 30-minute interview podcast (illustrative)

The following is an illustrative example, not a client project, using realistic figures of the kind you would see on a two-person remote interview recorded on decent USB microphones. The delivery target is a hypothetical podcast specification of -16 LUFS integrated with a -1 dBTP true-peak ceiling.

Look at where the work happened. Most of the level change came from clip gain and gentle per-voice compression. The limiter added only the last 1.5 dB. If the same target had been reached by pushing the raw -27 LUFS mix 11 dB into a limiter, laughs, plosives and every loud consonant would have been flattened, and the room tone would have risen along with everything else.

Deliverables checklist

• Integrated loudness within the specification's tolerance, measured over the whole program.
• True peak at or below the ceiling, measured with a true-peak meter, not a sample-peak meter.
• Loudness range appropriate to the content and playback environment.
• A quick check of the encoded file (for example the MP3 or AAC actually delivered), since lossy encoding can raise peaks.
• Session notes recording each dynamics stage and its purpose.

Short-form social video is a special case, because it is often heard on phone speakers with the volume low. It generally benefits from a tighter loudness range, but the transient and matched-comparison rules still apply. Our practical guide to audio for short-form video covers those platform-specific choices, and our audio sweetening service handles dynamics, EQ and loudness compliance for finished video.

Knowing When the Dynamics Need a Second Pair of Ears

Most of this checklist can be applied by anyone with a DAW, a loudness meter and the patience to listen carefully. There are three situations where bringing in help usually pays for itself. The first is a mix that sounds flat and lifeless after processing: the level is steady, but the energy has gone, and adding more EQ or more compression only makes it worse. That typically means transients have been lost at several stages and the chain needs rebuilding rather than adjusting. The second is dialogue that pumps audibly, where background noise swells between words or the voice seems to breathe. That is a release, detection or noise-floor problem, often compounded by aggressive noise reduction, and it is hard to fix by trial and error. The third is a master that must meet a loudness specification without over-compression, particularly for broadcast delivery, where a rejected file costs time and money.

An experienced engineer will approach any of these by asking the first question on this checklist: what is each processor for? From there, they will usually remove processors before adding any, compare at matched loudness and rebuild the chain in gentle stages. If you are evaluating outside help, our guide to choosing an audio studio covers what to ask, and our audio editing and production services include dynamics, repair and loudness-compliant mastering for spoken word, video and music.

Where this comes from

• Audio Engineering Society — Dynamic range processing
• iZotope — Dynamics documentation
• Steinberg — Mixing and processing guides

The figures and practices above come from the sources listed.

Working on something like this?

We take on Audio Editing & Production work for teams who want it done once, properly. Tell us what you are building and we will tell you honestly whether we are the right studio for it. Start a project.

Where to go next

• Audio Branding and Sonic Identity: What Actually Works
• Remote Recording and Double-Enders: The Decisions That Matter