Noise is far easier to prevent than to remove. Every decibel of hiss, hum or rumble you capture gets louder when the voice is compressed and raised to a delivery level such as -16 LUFS for a podcast or the -23 to -18 dB RMS window ACX requires for audiobooks. Processing can reduce steady noise, but it always trades away some voice quality to do it. This guide starts with the fixes that cost nothing at the recording stage and ends with what noise reduction realistically can and cannot do.
Know what you are fighting
Different noises need different fixes. It helps to sort them into three groups:
- Steady noise: hiss from a preamp, computer fans, air conditioning, a fridge, or electrical hum. It is constant, which makes it the easiest kind to reduce later.
- Intermittent noise: traffic, dogs, doors, footsteps upstairs, notifications, a creaking chair. It comes and goes, and processing handles it poorly.
- Room sound: reflections from bare walls, desks and windows. It is not noise in the usual sense, but it makes a voice sound distant, and a noise reducer cannot remove it.
Before you change anything, record 20 seconds of silence in your normal setup with your normal gain, then listen back on headphones at a high volume. Measure it with the free ACX check, which reports the noise floor as the RMS level of the quietest stretch. That number is your baseline for every change that follows.
Choose and treat the room
The room matters more than the microphone. A small room full of soft furnishings, such as a bedroom with a bed, curtains and a full wardrobe, usually sounds better than a large empty room with hard walls. Some practical steps:
- Record away from windows and exterior walls, which let in traffic noise.
- Face the softest surfaces in the room, such as a sofa, curtains or a full bookshelf, rather than a bare wall close to your face.
- Put something soft on the desk in front of you. A hard desk reflects sound straight back into the microphone.
- Close doors, and if you can, hang a heavy blanket or duvet over them.
- Record at the quietest time of day for your building, and keep to it for long projects so sessions match.
Foam panels help with reflections, not with noise coming from outside. Keeping outside noise out takes mass and sealed gaps, which is a building problem rather than a furnishing one.
Get closer to the microphone
The single most effective noise fix is distance. Each time you halve the distance between your mouth and the microphone, the direct sound of your voice rises by roughly 6 dB, while the background noise in the room stays where it was. You need less gain to reach the same recording level, and everything else in the room is captured lower relative to your voice.
Close recording has side effects. Directional microphones gain bass as you get closer (the proximity effect), and plosives such as "p" and "b" hit the capsule harder. A pop filter or foam windscreen, and speaking slightly across the microphone rather than straight into it, deal with both. A consistent distance matters as much as a short one: if you lean back and forward, both level and tone change.
The type of microphone plays a part too. A dynamic cardioid microphone used close up picks up less of the room than a sensitive condenser placed further away. Whatever you use, point its least sensitive side (the back, for a cardioid) toward the main noise source, such as a computer or a door.
Set your gain properly
Gain is often blamed for noise, but raising the input gain lifts the room noise and your voice by the same amount, so it does not change the ratio between them. What gain does affect is the noise of the equipment itself. If you record very quietly and boost later, the hiss of the preamp and converter is boosted with you.
- Set the gain so your loudest natural speech stays comfortably below clipping, with room for laughs and emphasis.
- Record at 24-bit if you can. It lets you leave generous headroom without pushing quiet passages down toward the converter's own noise.
- If you are near the top of the gain range, moving closer or using a more sensitive microphone is usually a better fix than more gain.
- Turn off automatic gain control in your recording app or microphone software. It raises the background noise in every pause.
Silence the building
A lot of background noise comes from things you have stopped noticing:
- Heating, ventilation and air conditioning. Switch it off during takes if you can, and back on between them. Moving air also produces low rumble that the ear tunes out but the microphone does not.
- Fridges and freezers. Their compressors cycle on and off, so the noise changes in the middle of a take. Record away from the kitchen, or pause until the cycle ends.
- Computers. Fans spin up under load. Move the computer away from the microphone or under the desk, close heavy applications, and point the microphone's dead side at it.
- Phones. Use airplane mode, not just silent. Notifications and vibration are obvious, but a phone close to a cable or interface can also cause buzzing interference.
- Electrical hum. A steady hum at the mains frequency (50 or 60 Hz, depending on your country) and its multiples often comes from ground loops, dimmer switches, or audio cables running alongside power cables. Plug the computer and interface into the same power strip, keep audio cables away from power supplies, and try a different USB port.
- Movement. Rustling clothes, jewellery, a creaking chair and tapping the desk all end up in the recording. Mount the microphone on a stand or arm instead of standing it on the desk.
Record room tone
At the start of each session, record a few seconds of the room with nobody speaking. You will use it to fill edits, so cuts do not drop into absolute silence, and noise reduction tools can learn the noise profile from it. For audiobooks, ACX recommends 1 to 5 seconds of room tone at the beginning and end of every file, and it should be real room tone, not digital silence. If you record audiobooks, getting the noise floor below -60 dB goes into the ACX side in detail.
What processing can do
Once the recording is as clean as you can make it, processing can take care of what remains, within limits.
High-pass filtering
A high-pass filter removes energy below a set frequency. For speech, a cutoff between about 60 and 100 Hz, often 80 Hz, removes rumble from traffic, footsteps and air conditioning without thinning the voice; thin or high voices can go a little higher. Rumble is worth removing even when you cannot hear it on small speakers, because it uses up headroom and raises peak levels.
Spectral noise reduction
Spectral noise reduction splits the audio into many frequency bands, estimates the noise level in each, and turns down the bands where the signal is close to that noise. It works well on steady noise such as hiss, hum, fans and air conditioning. Modest amounts of reduction tend to be transparent; large amounts start to change the voice.
Expanders and gates
A downward expander lowers the level when the voice stops, so the noise between phrases drops. It does nothing for the noise underneath the words, and an aggressive gate that cuts pauses to silence makes the background audibly switch on and off with the voice.
What processing cannot do
- Intermittent sounds. A bark, a horn or a door slam over speech is not steady, so a noise profile cannot describe it. Edit the moment out or re-record the line.
- Room echo. Reflections are delayed copies of your own voice, not noise. A noise reducer leaves them in place, and pushing it harder only damages the voice.
- Noise as loud as the voice. When noise and speech share the same frequencies at similar levels, any reduction removes voice as well.
- Heavy reduction without artifacts. Strong settings produce a watery or metallic sound, swallow word endings and soft consonants, and can leave a warbling residue often called musical noise.
Check noise reduction on headphones. Artifacts that are obvious on headphones are easy to miss on laptop speakers, and they cannot be taken out of the file afterwards.
Loudness processing also changes how noise sounds. Compression and levelling raise quiet material, including the room, so a background that was acceptable in the raw file can become distracting in the master. LUFS vs RMS vs peak explains how those levels are measured.
LufsReady's processing follows the same order as this guide: an 80 Hz rumble filter, spectral noise reduction set to off, light (6 dB), medium (12 dB) or strong (20 dB), then a de-esser and a leveler before the loudness is set. Like any noise reducer, it handles steady noise well and will not rescue a bark or an echoey room, which is why the recording fixes above come first. How it works describes each stage.
Sources
- ACX: Audio Submission Requirements
- Audacity manual: High-pass filter
- Audacity: Analysis plugins (ACX Check)
Check your file in seconds
Measure loudness, true peak and noise floor for free in your browser, or master a file to the exact target with a report.