How do EQ and compression make a voice sound better?

EQ fixes which frequencies are wrong, a compressor fixes which moments are too loud or too quiet. Starting values for speech, a worked example, signs of overdoing it, and where each tool actually runs on a Mac.


Record thirty seconds of yourself reading a paragraph, then play it back on headphones. Most people hear the same four things: a low rumble under every word that nobody noticed in the room; a boomy, slightly stuffed tone, as if the voice came through a cardboard tube; the ends of sentences sinking into nothing; and one laugh that jumps out so far you reach for the volume.

None of that needs a new microphone. The first two are tone problems, and an equalizer fixes tone. The last two are level problems, and a compressor fixes level. Both are standard on broadcast voices, and they are much less mysterious than their control panels suggest.

Below: what each one does, starting values for speech you can type in today, how to tell you have gone too far, and where you can actually run them on a Mac. If you would rather skip the knobs and just want to know whether a downloaded preset can do the job, read Do vocal presets work, or does your voice need its own settings? instead; it covers why a preset tuned on someone else's voice and mic often misses. And if your problem is broader than tone and level (noise, distance, a bad room), start with how to make your mic sound better without buying a new one.

Two tools, two different questions

The quickest way to keep them apart: an EQ answers "which frequencies are too loud or too quiet?", a compressor answers "which moments are too loud or too quiet?".

Equalizer (EQ)Compressor
Works onFrequency: bass, mids, trebleLevel over time: loud and quiet moments
FixesBoom, mud, nasal honk, dullness, harshnessWords that drop away, laughs that spike, uneven delivery
Main controlsFrequency, gain, width (Q), filter typeThreshold, ratio, attack, release, output gain
Typical mistakeBoosting everything that sounds "better" for ten secondsSquashing so hard that breaths and room noise come up
You notice it asThe voice sounds closer, cleaner, less boxyThe voice sits at one steady, easy-to-follow volume

They work together because each covers the other's blind spot. An EQ cannot stop you getting louder when you get excited. A compressor cannot remove the boom from a mic placed too close. You usually want both, in moderation.

Vocal EQ: what each part of the spectrum does to a voice

A voice is not one frequency. The fundamental of a speaking voice sits low, but the parts that make words understandable live much higher, where consonants are. iZotope's guide to EQ for vocals splits the range into bands that are a useful map for speech too:

RangeWhat lives thereToo much soundsToo little soundsUsual move for speech
Below 50-100 HzRumble, desk bumps, handling noiseMuddy, "thumpy"(nothing useful lost)High-pass filter
100-400 HzBody and warmthBoomy, muddyThinSmall cut if boomy
400-800 HzBoxinessPapery, boxyHollowNarrow cut if boxy
800 Hz-1.5 kHzNasal resonanceHonky, nasalDistantNarrow cut, only if needed
1.5-5 kHzPresence, intelligibilityHarsh, tiringFlat, muffledBoost or cut of 1-2 dB at most
5-8 kHzSibilance ("s", "t")Hissy, spittyLispyDe-esser, not broad EQ
8-12 kHzSparkleBrittleDullGentle boost if dark
12-20 kHzAirHissyClosed-inOptional small shelf

The iZotope guide notes that our ears are most sensitive around 2-5 kHz and that moves there of no more than 1 or 2 dB are common. A 2 dB change in that range is easy to hear; 6 dB is usually too much.

Start with a high-pass filter

A high-pass filter (also called a low cut) removes everything below a set frequency. For speech it is the single most useful EQ move because the lowest octave carries no words at all, only desk thumps, air conditioning, traffic and the boom you get when you lean close to the mic. Set it somewhere between 80 and 100 Hz for most voices. A deep male voice may want it nearer 70 Hz; a high voice can go to 120 Hz without losing anything. If the voice starts to sound thin, you set it too high.

Cut before you boost

A clean voice and a harsh one usually differ in one habit: fixing problems by removing them instead of piling energy on top. If the voice is boomy, cut 2-3 dB somewhere in 200-400 Hz. To find the exact spot, make a narrow boost of 8-10 dB, sweep it slowly through the range until the boom jumps out, then flip that same band into a cut. The sweep finds a resonance in seconds.

Boosts are fine, but small. A wide boost of 1.5-2 dB around 3 kHz makes most voices clearer on a laptop speaker. If you need more than that, the problem is almost always mic distance or mic position, not EQ. A mic too far away sounds thin and roomy; one too close sounds boomy because of the proximity effect, and moving it 5 cm fixes more than any filter.

Keep "s" sounds out of the EQ

Sharp "s" and "t" sounds sit around 5-8 kHz. Cutting that whole region with an EQ makes the voice dull all the time to fix a problem that lasts a tenth of a second. A de-esser is a compressor that only listens to that band, so it turns down the "s" only when it happens. If your editor has one, use it; if not, leave the band alone and avoid boosting there.

What does a compressor do?

A compressor is an automatic hand on the volume fader. When your voice gets louder than a level you choose, it turns it down by a proportion you choose. When you get quiet again, it lets go. After that, you raise the whole track back up, and the quiet words are now closer in level to the loud ones.

The OBS knowledge base puts it in one line: a compressor "makes loud sounds quieter". Five controls decide how.

Threshold. The level where the compressor starts working. Anything below passes untouched. Set it too high and nothing happens; set it too low and every syllable gets pressed down, including breaths.

Ratio. How hard it pushes once the voice crosses the threshold. OBS documents the arithmetic: at 2:1, a signal 6 dB above the threshold comes out only 3 dB above it, and OBS calls 2:1 weak and 6:1 much stronger. The same math at 3:1: a word 9 dB over the threshold comes out 3 dB over, so the compressor took 6 dB off.

Attack. How fast it reacts, in milliseconds. Very fast attack catches the first spike of a "p" or "k" and can make speech sound flat. A slightly slower attack lets the start of each word through, which keeps consonants crisp.

Release. How fast it lets go once you get quieter. Too fast, and the level flutters between words. Too slow, and the voice stays turned down after a loud word, so the next quiet one disappears.

Output gain (make-up gain). Compression makes the overall level lower, so you add gain at the end to bring it back. This is the step that makes quiet words louder. A standard (downward) compressor, the kind discussed here, only ever turns things down; OBS also has a separate Upward Compressor filter, which works the other way.

Some compressors also have a knee, which decides whether compression starts abruptly at the threshold (hard knee) or eases in around it (soft knee). For speech, a soft knee is the forgiving choice.

A worked example

Say your normal talking voice peaks around -18 dB on the meter, your quieter trailing words sit near -26 dB, and a laugh hits -6 dB. With the threshold at -20 dB and a ratio of 3:1:

  • normal speech at -18 dB is 2 dB over and comes out at about -19.3 dB (barely 1.3 dB of reduction);
  • the laugh at -6 dB is 14 dB over and comes out about 4.7 dB over, at roughly -15.3 dB, so it lost more than 9 dB;
  • the trailing words at -26 dB are untouched.

Now add 5 dB of output gain. Normal speech lands near -14 dB, trailing words at -21 dB, the laugh around -10 dB. The gap between the quietest and loudest moments shrank from 20 dB to about 11 dB. Same voice, same mic, far easier to listen to.

Compressor for vocals: starting settings for speech

Default settings are often wrong for talking. OBS ships its compressor at 10:1 with a -18 dB threshold, which is strong for speech: almost every syllable gets pressed hard. Here is a gentler starting chain for a spoken voice, in order:

1. High-pass filter     80-100 Hz, 12-24 dB per octave
2. Corrective EQ cut    -2 to -3 dB around 200-400 Hz, only if boomy
3. Compressor           ratio 3:1
                        threshold: set so normal speech shows 2-4 dB
                        of gain reduction, loud words 6 dB or so
                        attack 10-20 ms
                        release 80-150 ms
                        soft knee if available
                        output gain: add back what the meter shows
4. Tone EQ boost        +1.5 to +2 dB, wide, around 3 kHz
5. Limiter              ceiling -1 dB, catches whatever slips through

Treat those numbers as a place to start and then listen. If your compressor shows a gain reduction meter, watch it: if it sits at 10 dB all the time, the threshold is too low or the ratio too high. For a calm, consistent talker, 2:1 may be plenty. For someone who swings from mumbling to shouting, 4:1 with a slower release can work better.

EQ before or after compression?

Put the cleanup EQ (high-pass and corrective cuts) before the compressor. If boom and rumble reach the compressor first, it reacts to energy nobody wants to hear and turns your words down because of it. Put tonal boosts after the compressor, so the compressor is not fighting the extra presence you added. A limiter, if you use one, always goes last.

If you are building this chain inside OBS, its filter list runs top to bottom, and the OBS mic filters guide walks through where a noise gate, noise suppression, compressor and limiter each belong. The OBS knowledge base suggests placing the compressor at or near the beginning of the chain, which fits: only the cleanup goes in front of it. For noise gate values specifically, see the right noise gate settings in OBS.

Podcast EQ and loudness: the number at the end

For a podcast, compression is half of the story; the other half is how loud the finished file is. Apple Podcasts recommends an overall loudness of around -16 dB LKFS, with a 1 dB tolerance, and true peaks that do not exceed -1 dB FS. The same page warns that content that is heavily compressed and amplified can end up too loud and lacking dynamic range.

That gives a sensible order for podcast EQ and processing: clean up with EQ, smooth with moderate compression, then use a loudness normalization step to land on the target. Audacity's Loudness Normalization effect, based on the EBU R 128 standard, does exactly that last step. The common mistake is pushing the compressor harder to hit the number. The compressor evens out the voice; the normalizer sets the level.

How to tell you overdid it

Over-processing rarely sounds bad in the first ten seconds. It sounds bad twenty minutes into a call or an episode. Check for these symptoms on a headphone playback:

What you hearLikely causeFix
Every breath is as loud as a wordToo much gain reduction plus a lot of output gainRaise the threshold, lower the ratio
Room noise swells up in the pausesRelease too fast, threshold too lowSlower release, less make-up gain, consider a gate
Voice sounds flat, lifeless, "pressed"Attack too fast, ratio too highAttack to 15-20 ms, ratio down to 2:1 or 3:1
Harsh, tiring after a few minutesPresence boost too bigKeep it under 2 dB, or cut instead of boost
Thin, small, tinnyHigh-pass set too highBring it down toward 80 Hz
Still boomyMic too close, not an EQ problemBack off a few centimeters, then cut 200-400 Hz
Sharp "s" sounds hurtTreble boost or bright micDe-esser, or remove the boost

The Audacity manual says the same thing in its own words about vocals: pressing down the loud parts makes breathing and other close-up sounds very audible. That is not a bug. It is how compression works, and it is why less is usually more on speech.

Where you can run EQ and compression on a Mac

This is where Mac users get stuck. The tools exist, but each one works only inside its own app.

For recordings. GarageBand is free and has an eight-band EQ under the Smart Controls button, including a high pass filter and an Analyzer that shows the frequency curve live as you play; a Compressor can be added to any empty plug-in slot. The GarageBand voice recording guide covers the whole setup. Audacity has Filter Curve EQ and Graphic EQ under Effect > EQ and Filters, and a Compressor under Effect > Volume and Compression that can run in realtime on a track. If Audacity is your editor, see setting up Audacity on a Mac.

For streams. OBS has a Compressor, a Limiter and, since version 29, a 3-band equalizer filter. Those apply to what OBS records and streams, not to your Zoom call happening at the same time.

For live calls. This is the gap. On a Mac, Zoom's documented audio settings give you input volume, automatic volume adjustment and microphone modes for noise removal, and no equalizer. On the Mac itself, System Settings > Sound > Input gives you a device picker and an input level slider. To put an EQ and a compressor on your voice in Zoom, Meet, Teams or Discord, the processing has to happen before the call app sees the mic, which means a virtual microphone. One way is Rogue Amoeba's Audio Hijack, whose blocks include a Simple Compressor and a Parametric EQ, paired with Loopback, which Rogue Amoeba describes as the way to send the adjusted audio on to another app; Audio Hijack or Loopback explains how those two split the work. That route gives full manual control, and you still choose every value yourself.

Where TunedMic fits, and where it does not

For live calls, TunedMic works on tone, clarity and loudness, with one difference: you do not pick the values from a table written for an average voice. TunedMic adds a virtual microphone called TunedMic to your Mac. Its setup, Sound Check, asks you to read a short passage aloud for about a minute while it takes more than 30 measurements of your voice, your mic and your room. The profiles it then offers (Natural, Warm and Focused for calls; Balanced, Broadcast and Crisp for streams and podcasts; three more for singing) are built from that Sound Check, not from a stock preset.

You pick TunedMic as the microphone in Zoom, Meet, Teams, Slack, Telegram, Discord or OBS, and every one of them gets the processed voice at once. If you want to steer it, there are optional manual controls for noise, room echo, clarity, loudness and a 12-band tone equalizer; you never have to touch them. It adds less than a tenth of a second of delay and runs locally, so the voice is never uploaded.

The honest limits. TunedMic needs a Mac with Apple silicon (M1 or newer) and macOS 14 or later; there is no Windows or Intel version. It does not replace an editor: if you are mixing a podcast episode with music and two guests, GarageBand, Audacity or a DAW is still the right place. And processing cannot undo a mic held at arm's length in an echoey kitchen; fix distance first.

Your next step

Do the thirty-second test from the top of this page tonight. Record yourself in GarageBand or Audacity, add a high-pass filter at 90 Hz and a compressor at 3:1 with the threshold set for about 3 dB of reduction on normal speech, add the output gain back, and compare the two versions on headphones. If you hear the difference and want the same effect on every call without building a chain, watch the 38-second video on the home page: it shows the one-minute Sound Check doing that tuning for you and TunedMic picked as the mic in Zoom, Google Meet and Discord. TunedMic has a free 7-day trial that starts on first launch, with every feature and no card or account; after that it is $19 once, as listed on pricing. There are no refunds, so use the trial on your own calls before you decide.

Keep reading