Noise reduction
The hum of a computer or the rumble of the street can ruin a good conversation. Noise reduction helps a lot, but it has a price: overdo it and the voice turns metallic. Learn to clean up without wrecking things.
First of all: the best noise is the noise you don't record
No program can give you a truly clean recording from a dirty one. All noise reduction removes some voice along with the noise. That's why your first tools are the room and mic technique: turn off the fridge, move away from the computer and speak close to the mic. The louder your voice is relative to the background, the less reduction you'll need and the better the result will sound.
Constant noise and one-off noises
Not all noise is treated the same. The first step is to identify what you've got.
| Type of noise | Examples | How to treat it |
|---|---|---|
| Constant broadband | Preamp hiss, fan, air conditioning | Profile-based or AI noise reduction |
| Electrical hum | A fixed low tone at 60 Hz (mains power in North America; 50 Hz in the UK and Europe) and its multiples | A dedicated hum filter (de-hum) or very narrow filters |
| One-off noises | Cough, bump on the desk, car horn, dog | Cut, turn down or fill with room tone |
| Clicks and pops | Mouth noises, saliva, rubbing | De-click tools or editing by hand |
| Reverb | Empty room, “sounds like a bathroom” | De-reverb tools, with limited results |
Constant noises are the easiest to reduce, because the program can learn what they're like. One-off noises are better fixed by cutting and cleaning by hand.
The noise profile and room tone
Classic noise reducers work in two steps. First you select a section where there's only noise, no voice, and the program creates a noise profile: a “fingerprint” of which frequencies it occupies. Then you apply the reduction to the whole audio and the program subtracts that fingerprint.
That's why recording room tone is so useful: 20–30 seconds of silence in the room, with everyone quiet and in position, at the start or end of the session. That snippet is the perfect profile. If you don't have it, look for the longest, cleanest pause in the episode.
Every microphone and every room has its own noise. If you have several tracks, create a profile for each. And if you record on another day with a different setup, make a new profile: last week's won't work.
How much to reduce: the sweet spot
This is the key setting. Gentle reduction lowers the background without being noticeable; aggressive reduction removes it but leaves artifacts: the voice sounds metallic, as if speaking inside a can or underwater, and little “warbles” appear in the background. Move the reduction in the simulator above and you'll hear the moment the voice starts to break down.
- Start with moderate reduction, around
6–12 dB, and increase it gradually. - It's better to lower the noise than to eliminate it: a soft, steady background is far less annoying than a voice full of artifacts.
- If you need a lot of reduction, try two gentle passes instead of one heavy one.
- Listen on headphones and compare by switching the effect on and off.
Don't
Reduction at maximum: the background disappears, but the voice sounds metallic and bubbly.
Do
Moderate reduction: a very soft background remains and the voice keeps its timbre.
The noise gate
A noise gate doesn't clean noise while you speak: what it does is lower or mute the track when the level drops below a threshold, in other words during pauses. It's useful when several people have microphones, so that the mic of whoever isn't speaking doesn't add noise and room sound.
- Threshold: set it above the background noise and below the softest voice.
- Range: it's better to lower by about 10–20 dB than to mute completely; total silence between sentences sounds artificial.
- Attack and release: a fast attack so you don't clip the start of words, and a gentle release so you don't cut off the endings.
If the threshold is too high, the gate closes on soft syllables and chops off the end of words; if the background is noisy, you hear the noise popping in and out as people talk. On individual tracks it's often cleaner to edit by hand or turn down the pauses.
AI tools: powerful, but with limits
AI-based voice cleaners don't subtract a profile; instead they try to separate the voice from everything else. They work very well with variable noise, like traffic or background conversations, and they also reduce reverb. There are versions built into editors and online services.
Their limits: on strong settings they can make the voice sound artificial or overly “studio”, change the timbre, swallow consonants or breaths and, occasionally, invent sounds where the signal was very poor. If the platform is online, also think about the privacy of the content you upload. Use them at just the right intensity, compare with the original and don't apply them out of habit to audio that's already clean.
Order in the chain
Noise reduction goes at the start of the effects chain, after cutting and before EQ and compression. The reason: the compressor reduces the difference between the voice and the pauses, so it raises the background noise. If you clean it first, the compressor works on a cleaner signal.
- Avoiding noise when recording works better than any effect.
- Tell constant noise (noise reducer) from one-off noises (edit by hand).
- Record room tone so you have a good noise profile.
- Reduce in moderation: a soft background is better than a metallic voice.
- Gates and AI help, but set them by ear and always compare with the original.
Your turn: find the limit
Record a minute of voice with a constant background noise (a fan, for example) and find the sweet spot for reduction.
- Record 20 seconds of room tone and then a minute of talking with the fan on.
- Create the noise profile and export three versions: gentle, medium and maximum reduction.
- Listen to them blind on headphones and keep the one that best balances a clean background and a natural voice.
Frequently asked questions
How do I remove background noise from audio?
Select a section with only noise, create a noise profile and apply moderate reduction to the whole audio. You can also use an AI voice cleaner, without turning the intensity all the way up.
Why does my voice sound metallic after removing noise?
Because the reduction is too strong and creates artifacts. Lower the amount of reduction, or do two gentle passes instead of one aggressive one.
What is a noise gate for?
It lowers or mutes a track when its level drops below a threshold, during pauses. It's useful with several microphones so that the ones not in use don't add noise.