Noise gate vs noise suppression

Search for a fix for mic noise and you will find both tools recommended interchangeably, as if they were two brands of the same thing. They are not. A noise gate and noise suppression attack different halves of the same problem, and the fastest way to see that is to run one noisy recording through each and compare what comes out.

Same input, two different outputs

Picture a typical noisy mic signal: your words, with a steady carpet of hiss underneath them, and that same carpet filling the gaps whenever you pause. Fan noise, computer hum, the room itself. Now process it twice.

Through the gate: the gaps go dead silent. The gate watches your level, and once it drops below a threshold you set and stays there past a short hold, the channel fades to silence. But listen closely while you are talking and the hiss is still there, riding along under every word. The gate judges loudness and nothing else; while your voice holds it open, everything underneath your voice comes through too.

Through suppression: the hiss is turned far down everywhere, including under your words while you speak. Suppression separates voice from noise across the whole signal, not just the quiet parts. But it does nothing extra for the gaps: with the hiss turned down, your pauses carry faint residual room instead of the hard silence a gate delivers.

The two nodes' own displays in Patchd show why. Catch them at the same moment, in the middle of a sentence: the gate draws a level and a line, and suppression draws a voice detector.

So the answer to "gate or suppression?" is usually "yes." Suppression strips the floor, the gate seals the gaps. Partners, not rivals, and in Patchd both are free effects on every channel.

Where each one fails, and the setting that fixes it

Each tool has a characteristic failure mode, and each ships with a control built specifically to handle it.

• The gate chatters near the threshold. Trail off at the end of a sentence and your level hovers right at the line, flicking the gate open and shut like a bad relay. The fix is hysteresis (default 0 dB, up to 24): the gate opens at your threshold but refuses to close until you are that many dB quieter, so hovering between the two lines never flickers it.
• The gate clips the front of words. A gate is closed by definition when you start talking, so a slow attack shaves the first consonant off every sentence. Keep it fast: the default is 1 ms, and listeners forgive hiss long before they forgive missing consonants.
• Suppression can color your voice. Pushed hard, separating voice from noise starts to cost some of the voice: heavy settings can leave you sounding thin or processed. Patchd's suppression ships with an honest Mix control instead of a single on switch, so you blend between the raw and fully cleaned signal and stop where the noise is gone but you still sound like yourself.

Notice the asymmetry: the gate's failures are timing problems, fixed with more nuanced timing rules. Suppression's failure is a quality tradeoff, fixed by dialing back how much of it you take. That difference is not an accident, and the advanced tier explains why.

A threshold versus an estimator

Under the hood, the two tools are different kinds of machine. The gate reduces your entire signal to one number and compares it to a line. That number is the envelope, a running estimate of your level that grabs peaks instantly and lets go of them slowly:

Every decision the gate makes is that envelope against a threshold, both converted to linear amplitude. With hysteresis in play there are two lines, one to open and a lower one to close:

That is the whole gate: a one-dimensional judgment, made thousands of times per second. It is why the gate is cheap, predictable, and utterly blind to what the sound is. A cough and a word look identical to an envelope follower; only their loudness differs.

Suppression is the opposite kind of machine. Instead of comparing one number to a line, it runs a trained model that continuously estimates which parts of the incoming signal are voice and which are noise, and rebuilds the signal with the noise scaled away. It never asks "is this loud enough?" It asks "how much of this is you?", everywhere, all the time, which is exactly why it can reach under your words where a loudness test cannot. The cost is that its mistakes are estimation errors, not timing errors, which is why its safety valve is a Mix blend rather than another millisecond knob.

Running them together, and which goes first

Use both. Suppression handles the noise under your voice while you are talking, the gate handles the gaps when you stop, and neither one does the other job. Which of the two runs first is a judgment call rather than a rule, and it is worth knowing what each way buys you.

Suppression first hands the gate a quieter floor. Suppression can take a room from around -50 dB down to around -70, so a threshold that merely worked before now sits far below your quietest speech with headroom to spare, instead of being wedged a few dB above the hiss where soft words get clipped.

The gate first means suppression only ever runs on your voice. The model never works on the room in the pauses, which is the material it was least trained for, at the cost of a gate that has to judge the raw floor and so sits closer to it.

Set the threshold against the live meter for whichever order you pick, and listen to the quiet ends of your words. The ordering logic for the rest of the chain, including the worked example of how a misplaced compressor wedges a gate permanently open, lives in the effect chain order guide.

In Patchd, both effects are free nodes in the rack on every channel: drop them both in, set the suppression Mix by ear, and set the gate threshold against the live meter. Whatever the chain produces is what every destination hears, from your headphones to your stream. You can try it yourself: Patchd is in open beta now, and if you clear the terms your invitation arrives straight away.