Noise gate for audio: how it works and how to set one

A noise gate watches the level of your signal and lets it through only while it is above a threshold you set. Below that, the gate closes and the output goes silent. That is the whole idea.

This makes a gate very different from noise reduction. Noise reduction analyses the sound and tries to subtract the unwanted part from underneath the voice. A gate does nothing of the kind: while you are speaking, every bit of the noise is still there, untouched. It simply removes the noise in the gaps where you are not speaking.

That is the key to using one well. A gate is for making silence properly silent, not for making a noisy recording clean.

Threshold sets the level at which the gate opens. This is the control that does most of the work, and it needs to sit above your room noise but comfortably below your quietest speech.

Attack is how quickly the gate opens once the threshold is crossed. Too slow and it eats the beginning of words, which is the most common complaint about gates and the reason speech through a badly set one sounds clipped.

Hold keeps the gate open for a set time after the signal drops back below the threshold. This is what stops it chattering open and closed during the natural dips inside a sentence.

Release is how gradually it closes once hold expires. A slow release fades the room out rather than cutting it, which sounds much more natural than an abrupt drop to digital silence.

Start with the threshold. Record yourself sitting silently for ten seconds, then speaking normally, then speaking as quietly as you ever would. Set the threshold between the loudest point of the silence and the quietest point of the speech. If those two overlap, no threshold will work and the gate is not your answer.

Set attack fast, in the region of one to ten milliseconds. Speech begins with consonants that carry a lot of the intelligibility, and a slow attack removes exactly those.

Set hold to a couple of hundred milliseconds and release to somewhere between 100 and 500 milliseconds. Longer values sound smoother; shorter values remove more room tone. Err towards longer, because an audible gate is worse than a little room tone.

Streaming and conferencing tools often present a gate with two thresholds rather than one: a level at which it opens and a lower level at which it closes. This is called hysteresis, and it exists to stop the gate flickering when your level hovers right at the boundary.

Set the open threshold where speech reliably crosses it and the close threshold a few decibels lower. The gap between them is what gives you stability, and setting them too close together reintroduces the chattering the design was meant to prevent.

A gate is worth having on a live stream because it stops your keyboard, your chair and your room being broadcast whenever you stop talking. It is one of several settings worth getting right before you go live, alongside your audio bitrate and sample rate.

If the noise is loud enough to hear underneath the voice, gating will not help. It only cleans the gaps, and the moment someone speaks the noise returns at full strength, which draws attention to it. That problem needs noise reduction, not a gate.

A gate also struggles with quiet, intimate delivery, where the speech itself sits close to the noise floor. And it will cut breaths, which some people want and others find unnatural, because speech without any breath at all sounds synthetic.

For recorded material rather than live, consider editing the gaps by hand instead. It takes longer but you decide exactly what goes and what stays, which no threshold can do for you.

Put the gate first, before compression. Gating a signal that has already been compressed is much harder, because compression has raised the quiet parts and narrowed the gap between your room tone and your voice, which is exactly the gap the threshold depends on.

Gate, then compress, then set your final level. In that order each stage is working on material the previous one has already tidied.