Processing: an even, clean voice

You've recorded clean, no clipping, silence at the start. Now ten minutes of processing, after which the voice sounds even and full. There are four steps, and the order matters.

The order

Noise reduction first. Then EQ. Then the compressor. Loudness last. The order matters because each step amplifies whatever the previous one left behind. If you compress first and clean afterwards, the compressor lifts the noise along with the quiet words, and you'll have to clean twice as hard. One exception: the low cut from step two can go first of all, before the noise reduction. Rumble and vibration get in the way of both the noise profile and the compressor.

These tools exist in desktop video editors and audio editors; mobile editors usually have a thinner set. The names vary, the job is the same.

Step 1. Noise reduction

The most precise way: the software takes a noise sample, that silence at the start of your recording, and subtracts it from the whole track. Select the silence, hit "get noise profile" or the equivalent, apply to the whole recording.

A noise profile isn't available everywhere. If your editor doesn't have one, you're left with a noise-reduction slider or an "enhance voice" button. Try the button first: recent versions often do this better than manual settings. If the voice sounds odd afterwards, switch it off and work the slider yourself. The silence sample is still useful here: it lets you hear exactly what you're removing.

The main rule: not all the way up. Set the strength to a third, half at most. If the voice went dull, ringing or "underwater", that's too much: the tool bit off part of the voice along with the noise. A faint background beats a robot.

Step 2. EQ

An equaliser changes the volume of individual frequencies. For a voice you need one move, and optionally a couple more.

The must-do: cut the low end. Everything below 80–100 hertz in a voice recording isn't voice, it's rumble, desk knocks, floor vibration, ventilation. The tool is called a high-pass filter or low cut, depending on the software. Set it around 80 hertz for a deep voice, around 100 for a higher one. You'll barely hear a difference in the voice, but the mud is gone.

Optional. If the voice booms or sounds boxy, dip the 200–400 hertz area slightly. If it's dull, lift 3–5 kilohertz slightly, that's where clarity lives. If it's harsh, dip 2–4 kilohertz. If the "s" and "sh" sounds hiss, you want a de-esser, and without one, dip 6–8 kilohertz a touch. "Slightly" means 2–3 dB. Big EQ moves make things worse.

Step 3. Compressor

The most confusing tool for a beginner, and it does a simple thing: it reduces the difference between quiet and loud words. It squashes the loud ones, after which you can bring the whole track up, and the quiet words become audible. The result is an even, full voice, the kind you hear on the radio.

There are many settings, you need two. Threshold: at what volume it starts squashing. Ratio: how hard. For a voice, start at 2:1 to 4:1, which in most programs is the "voice" or "speech" preset. A well-set compressor pulls the loudest words down by 3–6 dB. If breaths and noise between words got noticeably louder, the compression is too strong, back it off.

Tip. A compressor doesn't rescue clipping and doesn't remove echo. It only evens things out. If the recording is bad, the compressor makes it bad and even.

Step 4. Loudness

The last step: bring the track to a standard loudness. Don't count on the platforms. YouTube turns down audio that's too loud but won't bring quiet audio up. Other platforms behave differently and change their rules, so don't count on them to level it. A quiet video may stay quiet. So deliver it at the right level yourself.

A good target for social and video: integrated loudness around −14 LUFS, peaks no higher than −1 dBTP. LUFS is a measure of perceived loudness, and editors have a loudness normalisation function where you can type that number in. If yours doesn't, normalise peaks to −1 and compare by ear with other people's videos opened side by side.

If the video will have music, do the final normalisation on the finished mix, not on the voice alone. Keep the voice around −16 to −14 LUFS before mixing, or the whole thing ends up above the target once the music is added.

Check on headphones and on the phone speaker. The speaker shows you what most people will hear.

What not to do

Don't stack five effects in a row. Don't chase a "studio" sound through presets with reverb: you just spent a lesson removing echo. Apply the processing to the whole track at once, not to each clip separately, or the loudness will jump between clips. And after every step, switch the effect on and off at the same volume: the ear easily confuses "louder" with "better".

Do this now

• Open a recording in your editor, select the silence at the start and apply noise reduction at a third strength. Switch the effect on and off and compare.
• Set a low cut at 80–100 hertz and listen to what disappeared.
• Apply the voice compressor preset and normalise loudness to −14 LUFS. Open someone else's video next to it and compare the volume.