Once you have set levels, panned tracks, and balanced a mix, the remaining tool in the mixing toolbox is dynamics processing: changing the relationship between the loud and quiet parts of a sound. This reading looks at the three tools that do this work (normalization, compression, and limiting), what each one does, when each helps, and how the powerful ones get misused.
The dynamic range of a sound is the difference between its quietest and loudest parts, measured in decibels (dB). A whisper followed by a shout has a wide dynamic range. A car alarm has a narrow one: it's loud, it stays loud, the loud and quiet parts aren't very different.
Pieces of music have dynamic range too. A solo piano recording has a wide range: a soft passage might sit 30 or 40 dB below a loud chord. A modern pop song mastered for radio has a narrow range: from start to finish, everything sits within a few dB of the loudest peak.
Neither is automatically better. Wide dynamic range gives a piece room to breathe, and lets quiet moments feel quiet. Narrow dynamic range gives a piece consistent presence, useful in environments where the listener can't adjust their volume (in a car, on a phone speaker, on a noisy bus).
Listen to the same musical fragment with wide and narrow dynamic range:
Both files have the same peak level. The narrow version sounds louder on average because the quiet parts have been brought up. But you can hear what it's lost: the conga slaps no longer stand clearly above the shaker and clave; everything is closer to the same level than it was in the wide version.
Look at the two waveforms below. Both are stretched to fill the same vertical space, and the dashed line marks exactly the same height in each panel: that's the −3 dB peak ceiling. The conga slaps in both versions touch that line. Look how different the spaces between the slaps are.
Your peak meter only measures the loudest single moment, which is what the dashed line tracks. But your ears measure something different: how much sound energy is hitting them, averaged over the last few hundred milliseconds. We call this average level. The wide version has tall peaks and deep valleys, so its average is low. The narrow version has the same peaks but the valleys are filled in, so its average is much higher.
| Peak level | Average level | |
|---|---|---|
| Wide | −3 dB | −24 dB |
| Narrow | −3 dB | −18 dB |
Identical peaks. About 6 dB louder average. Six decibels is a lot: roughly the perceptual difference between a normal conversation and a loud conversation. Your peak meter sees no difference between these two files; your ears hear a huge difference.
This is the secret behind every "loud master" in commercial music. You don't make a piece louder by pushing the peaks higher; you can't, they're already at the ceiling. You make it louder by pulling the average level up toward the peaks. That's what compression and limiting do.
Two knobs, not one. Strictly speaking, the narrow version above is compressed and then boosted. Compression by itself reduces the loud parts of a signal, which actually makes the file quieter overall (the loud bits got smaller, the quiet bits didn't change). To get the louder result you just heard, we then boost the whole compressed signal so the peaks come back up to the ceiling.
This second step is so common that most compressor plugins have a dedicated knob for it, usually labeled makeup gain (sometimes "output" or "gain"). It's just a multiplier applied after the compressor does its work. The compressor squashes; makeup gain pulls the squashed version back up.
The compressor in an audio editor like Audacity and the interactive tool in this reading both have a makeup gain control. Sometimes you'll want it; sometimes you won't. Compression without makeup gain is useful when you just want to tame a sound that's too loud, leaving the quiet parts where they are. Compression with makeup gain is what makes a piece feel louder. We combined them here to make the loudness change unmistakable, but they're separable choices in practice.
Any piece you make has some dynamic range, whether you've thought about it or not. Some moments are louder, some are quieter, and that contrast is part of how the piece holds attention. Normalization, compression, and limiting are the tools you use to change that relationship deliberately. The next sections introduce them one at a time.
Before we get to the compressor, there's a much simpler tool worth knowing: normalization. You'll use it almost every time you prepare a sample to keep, and it's the most common technique for making a quiet recording louder.
Normalizing a sound does exactly one thing: it finds the loudest peak in the file, computes how much to multiply every sample by so that peak sits at a target level (of your choosing), and applies that single multiplier to every sample in the file. The operation is a pure rescaling: one number applied uniformly across the whole file. The whole signal goes up (or down) by the same amount of dB. The relationship between loud and quiet parts stays exactly the same.
This makes normalization fundamentally different from what you heard in Section 1. The compressed-and-boosted "narrow" version had its dynamic range squashed: the quiet parts came up more than the loud parts. Normalization can't do that. Normalization is just a scaler. If the loud parts go up by 12 dB, the quiet parts go up by 12 dB. The shape of the signal is preserved.
Listen for the difference. The conga loop at two normalization levels: a quiet version (peak around -18 dB, as if it had been recorded with a conservative gain setting) and the same recording normalized to a peak of -1 dB:
The normalized version is louder. That's the only difference. The relative heights of the conga slaps, shaker, and clave are exactly the same; the shape is exactly the same. This is what "the whole signal scaled by one number" actually sounds like.
Compare this to the wide/narrow comparison in Section 1. There, both files had identical peak levels but very different shapes. Here, the two files have very different peak levels but identical shapes. Normalization changes the scale. Compression changes the shape.
Yes. This is easy to miss: changing a signal's shape over time also changes its tone. When a compressor squashes the sharp start of a hit or thickens a sustained note, it is reshaping the waveform from moment to moment, and reshaping a waveform inherently changes what frequencies are present and at what levels. Two files can be at the same loudness and still sound very different.
Listen for it. Both clips below are at the same loudness (matched by RMS, the closest measurement we have to how loud the ear actually hears something). The first is the conga loop made louder by normalization (linear scaling). The second is the same loop made louder by compression and limiting (non-linear wave-shaping).
The compressed version has more presence in the upper-mids and high frequencies. The slaps feel a little thicker; the shaker and clave feel a little more forward. The two files measure the same loudness but they have a different tone. The compressor reshaped the waveform, and the reshaping pulled in spectral character that the scaled version doesn't have.
This is why "dynamics processing" is a slightly deceiving name for the category. Compression and limiting are sold as loudness tools, and they are. They are also color tools. Every compressor imparts some character to whatever passes through it, and different compressors have different characters.
This isn't only marketing. The basic idea of a compressor (detect when the signal is loud, reduce the gain) is the same everywhere. But the specifics vary: how the detection works, what shape the gain reduction follows, how quickly it engages, whether the compressor deliberately adds a touch of distortion as part of its sound. A clean digital compressor like the one in an audio editor such as Audacity does the math accurately and adds nothing else. A vintage hardware unit, or a software emulation of one, was built (or modeled) to add a specific character on top of the basic compression. People are paying for that character. They are also paying for the engineering hours spent measuring an old piece of hardware accurately enough to reproduce its quirks in software.
Choosing a compressor for a particular sound is partly about how aggressively it can reduce gain, but it is mostly about which kind of character you want, and how cleanly or how colorfully the compressor delivers it.
For now, the takeaway is simpler: compression is never purely a loudness tool. It always changes the tone of what you put through it. Sometimes the tone change is what you actually wanted; sometimes it is a side effect you want to minimize. Either way, when you reach for a compressor you are reaching for a tool that does two things at once. Treat it that way.
Normalization is the standardization step in sample preparation. When you record sounds for a sample library, you set your input gain conservatively so nothing clips. That's the right thing to do at the time of recording. But you'll end up with samples at all kinds of different peak levels: this one peaks at -10 dB, that one at -16, the loud one at -6. When you later drag these into a session, you'll have to ride the channel fader 10 dB up for one sample and 4 dB down for another, just to get them to a usable working level. That's friction you don't need.
The solution is to normalize every sample to a consistent target peak before you save it to the library. After normalization, every sample sits at the same predictable peak, and you can work with them without thinking about base level. Quiet sample, loud sample: they all hit your channel at the same height. The mix is now about your decisions about relative loudness, not about which microphone gain you happened to set three weeks ago.
You'll do this whenever you prep a sample library, standardizing the level of each sound.
Normalize in an audio editor like Audacity defaults to a target of 0 dB: the loudest possible level. For sample prep, this is too loud. A normalized sample at 0 dB has zero headroom: any additional processing (EQ boost, pitch shift, layering with another sample) can push it past the digital ceiling and clip.
For sample prep, normalize to -1 dB. This is common in production: it leaves a small margin so that inter-sample peaks (peaks that exist between sampled values, which can show up as clipping after digital-to-analog conversion) stay safe. You'll see -1 dB as the target in a lot of professional sample libraries and in delivery specifications for streaming platforms.
The menu path is Effect → Volume and Compression → Normalize. Two settings to know:
-1.0 dB.That's it. Apply, save your file. The peak now sits at -1 dB regardless of what level it had before.
If you look around the audio world, you'll see references to loudness normalization or LUFS (Loudness Units relative to Full Scale). This is a different kind of normalization that targets perceived loudness rather than peak level. Streaming platforms like Spotify, Apple Music, and YouTube use it to play all tracks at roughly the same apparent volume, regardless of how each was mastered.
LUFS measurement uses the same idea you learned in Section 1: integrate the signal's energy over time to get something close to how loud the human ear hears it, then scale the whole file so that measured loudness hits a target (typically -14 LUFS for streaming). It's why a loudly-mastered track and a quietly-mastered track sound about the same volume on Spotify, even though their peak levels can be very different.
For sample prep, peak normalization is the right tool, because you're standardizing individual sounds, not pieces. LUFS becomes important when you're finishing a full piece for delivery. For now, just know the term exists.
You've now heard what compression does. This section names the four controls that determine how it does it.
A compressor reduces dynamic range. It attenuates the loud parts of a signal while leaving the quiet parts alone, reshaping the waveform from moment to moment (which, as Section 2 showed, also affects tone). Two parameters define how much attenuation happens and where it starts: threshold and ratio. Two more, which the next section covers, define when the compression engages and disengages.
The threshold is a level (in dB) above which the compressor starts working. Signal below the threshold passes through untouched. Signal above the threshold is reduced.
The ratio determines how much the signal above the threshold is reduced. A ratio of 2:1 means that for every 2 dB the input goes above the threshold, the output only goes up by 1 dB. A ratio of 4:1 means 4 dB in gets 1 dB out. A ratio of 10:1 means 10 dB in gets 1 dB out: the loud signal is barely allowed to get any louder.
Listen to the same source through three compression settings. The source is six short percussive hits: the first and fourth at a peak level (-3 dB), the second and fifth quieter (-9 dB), and the third and sixth quieter still (-15 dB). In each setting, listen for which hits get compressed and which don't:
In the light setting, only the loudest hits get touched. In medium, the loud and middle hits are both compressed, and the quietest hits are now nearly the same level as the loud ones. In heavy, almost all the dynamic difference between hits is gone: they all sound roughly the same volume. All four files are normalized to the same peak level, so the perceived differences come from how the compression changed the relationship between hits, not from absolute level.
One more control shapes the corner where compression begins: the knee. A hard knee turns compression on abruptly at the threshold, the sharp angle in the curve above. A soft knee rounds that corner, easing the compression in across a few dB on either side of the threshold so the onset is gentler and harder to hear. Most compressors, Ableton's among them, offer both. Soft is the safer default for music; hard is for when you want the compressor to grab firmly and audibly.
Threshold and ratio describe how much compression happens. But sound is not static. It changes over time, sometimes very fast. Two more parameters control when the compressor engages and disengages: attack and release.
The attack is how quickly the compressor reacts when the signal crosses the threshold. A fast attack (1–5 ms) clamps down on the signal almost immediately. A slow attack (30–100 ms) lets the first part of a loud transient pass through before kicking in.
The release is how quickly the compressor stops working once the signal drops below the threshold. A fast release (50 ms or less) lets go quickly; useful for sounds with rapid changes, but can sound "pumpy" as the compressor rapidly opens and closes. A slow release (300+ ms) holds the gain reduction for longer, sounding smoother but potentially muting quiet parts that follow loud ones.
Why these matter: the most distinctive parts of many sounds are their transients, the sharp attack of a drum hit, the pluck of a string, the burst at the start of a spoken consonant. A fast attack can squash these transients flat, robbing the sound of its character. A slow attack lets the transient through and only compresses the body of the sound, usually a more natural result.
All four settings use the same threshold (-30 dB) and ratio (4:1): only the time constants differ. In fast attack, the leading edges of the kick and snare are noticeably softened, robbing the hits of their snap. In slow attack, the transients pop through while the body of each hit is held in check, usually the more natural-sounding choice. In fast release, you may hear the compressor "breathing" as it lets go quickly between hits. Slow release holds the gain reduction longer, giving a smoother, more consistent sound at the cost of some recovery time between events.
A useful starting habit: when in doubt, use a medium attack (10–20 ms) and a medium release (100–200 ms). These let transients through, sound natural on most material, and give you something to adjust from. Aggressive settings (very fast attack, very fast release) are powerful but easy to overdo.
A limiter is a compressor pushed to extremes: a very high ratio (often 20:1 or higher, effectively infinite) and a hard ceiling that the signal is not allowed to exceed. Everything you learned about compression still applies; a limiter is the same tool with the threshold and ratio cranked. It is used at the end of a mix to set a maximum level and, often, to push the perceived loudness of the piece up.
A limiter set with a ceiling of -0.3 dB will not let any sample exceed that level. Anything that would have gone louder gets clamped down. If you then add input gain before the limiter (structurally the mirror image of the makeup gain idea from Section 1, applied before the processor instead of after), the loud parts get clamped harder while the quiet parts ride up, and the piece becomes louder on average without ever exceeding the ceiling.
This is useful for delivering finished material to listening contexts that demand consistent level (a class listening folder, an online release, a streaming platform). It is also where the worst habit in modern audio production happens.
Starting in the 1990s and peaking in the early 2000s, commercial recordings competed to be the loudest. The reasoning was that louder records caught the ear when listeners scanned the radio dial, and that perception became an industry-wide arms race. Engineers pushed limiters harder and harder, crushing peaks to make the music feel louder on the listener's playback system.
The cost: lost dynamic range, squashed transients, distortion, listener fatigue. Streaming platforms have since added loudness normalization, which automatically lowers loud tracks to match a target level, meaning the loudness wars now lose you something (dynamic range and clarity) without even gaining the original benefit (sounding louder than the next track).
The lesson: limiting is a useful tool. Like every tool, it can be abused. A piece doesn't need to be the loudest possible to feel impactful; usually the opposite.
Listen to the same source at three levels of limiting. The peak level is identical across all three files: they all touch the same ceiling. But the average level climbs dramatically, and so does the audible damage:
Listen carefully: the peak meter on all three would read about the same. But the natural version has clear contrast between the conga slaps and the quieter shaker and clave underneath them. The light version is consistently louder on average, with the slaps still audible as accents but slightly tamed. The crushed version is loud everywhere: the slaps have lost their punch, the quiet moments have been pushed up to nearly the same level, and you can hear distortion on the transients. The piece has been made louder by being made worse.
Compression and limiting are powerful tools that are easy to overuse. A few useful principles:
One technique earns a mention because it sidesteps the main tradeoff. Parallel compression blends a heavily compressed copy of a sound underneath the untouched original. The squashed copy adds body and lifts the quiet detail, while the dry copy keeps its transients sharp. You get the density of hard compression without the flattened hits. It is a staple on drum buses, on vocals, and on a full mix, and the Read more below goes deeper.
Compression has a specific sound, or, more accurately, several specific sounds depending on the settings. Building a vocabulary for what compression does to a sound lets you hear when it's helping and when it's hurting. A few terms you'll hear in conversation:
These terms aren't required vocabulary, but they're useful shorthand. The deeper skill is the one underneath: hearing what a compressor is doing to a sound, and knowing whether that's what you wanted.
A compressor normally listens to the same signal it is squashing. Feed it a different signal as the trigger instead, through what the compressor calls its sidechain or key input, and one track decides how hard another gets turned down. The classic move: the kick drum triggers a compressor on the bass, so every kick ducks the bass for a moment and the two stop fighting over the same low end. Push it harder and that ducking becomes the rhythmic pump under most house, techno, and a lot of pop.
It is the pumping a compressor can fall into by accident, except here you cause it on purpose and tune it to the beat. In Ableton, the compressor has a Sidechain panel: switch it on, choose the source track, and set how much it ducks with the same threshold and ratio you already know.
Load a sound (or use the built-in demo) and apply compression with adjustable threshold, ratio, attack, and release. Watch the transfer curve change as you move the sliders. Watch the gain-reduction trace move with the waveform. Toggle the bypass to A/B against the unprocessed source. The point isn't to learn this specific tool; it's to develop ears and intuition for what each parameter does, so that when you reach for a compressor in any DAW you know what to listen for.
How to use it: Try the demo sound first if you've just opened the page. Press Play; let it loop. Move Threshold down. You'll see the transfer curve bend below threshold and the gain-reduction trace start to move. Try Ratio at 2:1 vs 10:1 to feel the difference between gentle and aggressive. Then try Attack very fast (1 ms) vs slow (40 ms) on percussive material: you'll hear transients get squashed, then released. Toggle Bypass to A/B against the unprocessed source. When you find a setting you like, bring up Makeup gain to compensate for the level you've lost.
The compressor here is a feed-forward digital compressor with a one-pole peak detector and a 2 dB soft knee, one of several standard architectures used in software compressors. The math is exposed in the visualization, but you don't need to know it to use the tool. Trust your ears.
If shaping how loud things sound is the part that grabbed you, there is a whole course in it. MUS 485, Advanced Sound Design I: Sampling, Editing, and Mixing, takes dynamics and the rest of mixing well past this first pass.
Sound on Sound on compression, from what the controls actually do to the techniques you build with them.
Using Your DAW’s Compressors & Limiters What the controls actually do Compressors How a compressor works Advanced Compression Techniques: Part 1 Gain-control elements Advanced Compression Techniques: Part 2 Mastering & multiband Parallel Compression Blend the compressed with the dry