How to Remove Background Noise From a Voice Recording

2026-11-14 ·

Background noise in a voice recording can be removed online, but not all of it and not in the same way. Steady mains hum and hiss come out almost cleanly when the software learns the noise from a few seconds of silence in the same file. Room echo, a keyboard click in the middle of a sentence or a neighbour's voice through the wall do not disappear that easily, and aggressive cleaning damages the voice itself. Below: what each kind of noise does, what to remove, what to leave, and how to record so that you rarely need to clean up at all.

Identify the noise first

Noise in voice recordings differs by source and by behaviour, and that decides whether noise reduction can handle it.

Type Typical source What it sounds like Can it be removed?
Mains hum Power adapters, chargers, LED lights, poor grounding A low, steady drone: 50 Hz in Europe and most of the world, 60 Hz in North America, plus harmonics at 100, 150 and 200 Hz Yes, with narrow notch filters
Hiss Preamp, audio interface, cheap electronics A broadband shhh that stays constant over time Yes, from a sample of silence
Room tone Fridge, ventilation, air conditioner, computer fans A low, even rumble, like the room breathing Partly: well, as long as it is not too loud
Echo Reflections from walls, ceiling and floor The voice sounds like a bathroom, with a tail Reduced with a separate tool; rarely removed fully
Clicks and thumps Keyboard, mouse, lips, breath on the capsule Short impulses in the middle of speech Point-by-point repair, sometimes with losses
Other voices, dogs, traffic Neighbours, street, other people Events with their own spectrum, unlike the room Poorly: cut out or re-recorded

Hum and hiss are steady. Their spectrum does not change, so noise reduction can learn it from a silent passage. Echo, clicks and other voices are unsteady: they appear in one place and not another, and one global filter cannot cope with them.

You can tell hum from room tone by ear: hum has a clear pitch and sits mostly in the low end, while room tone is a wide, even hush. On a spectrogram, mains hum shows as flat horizontal lines at 50, 100 and 150 Hz. Hum on a digitised cassette is a separate job: see de-hum for cassette restoration.

What cleans up well and what does not

Steady noise cleans up well: mains hum, hiss, an even room tone. The effect is immediate, and you can only hurt the voice by pushing too hard.

Echo sits in between. It is reverberation, not noise in the narrow sense, and ordinary noise reduction does not see it, so it needs a separate de-reverb step. The details are in remove reverb from audio online. In a live room echo can usually be reduced, but rarely removed completely.

Clicks and other voices clean up poorly. A click is a short impulse, and smearing it with a broad noise reducer leaves a hole in the speech. Repairing each click by hand works better. The same logic applies to vinyl crackle, described in Vinyl clicks: how to remove de-click. Another voice or a dog barking does not separate cleanly from your speaker; a damaged passage is usually easier to re-record or cut.

How not to ruin the voice

The most common mistake is pushing noise reduction to the maximum. Weak parts of the voice, such as consonants, breath and upper harmonics, get reduced together with the noise, and the speech turns dull. Tonal leftovers of the noise that the algorithm could not separate sound like bubbling or a metallic edge. That is where the "underwater" effect comes from.

What helps:

  • Clean in two or three gentle passes instead of one heavy one, and listen after each pass.
  • Keep a natural room tone. Absolute silence in the pauses sounds unnatural, and listeners feel it even without any terminology.
  • Compare with the original: switch between "before" and "after" on the same passage.
  • Do not normalise before cleaning. The noise rises with the voice and becomes harder to separate.
  • If you have the isolated voice track, do not clean a finished mix with music in it. Music throws off the estimate, because the noise reducer treats instruments as noise.

Step by step: cleaning up a voice recording

  1. Find a silent passage two or three seconds long. The noise reducer learns the spectrum of the noise from it.
  2. If there is mains hum, remove it first with a narrow filter at 50 or 60 Hz and its harmonics. Check that the voice keeps its tone.
  3. Reduce hiss and room tone with moderate noise reduction, using the silence as the reference. Stop when the noise stops bothering you, not when it disappears completely.
  4. Repair clicks individually. Fix short ones one by one; cut long thumps and replace them with a pause that has natural room tone.
  5. If you need to remove echo, do it as a separate step and compare it with the previous version. Pushing echo removal too far makes the voice sound hollow.
  6. Level the volume after cleaning, and do not raise a quiet file more than you need to: the leftover noise rises with the voice.

For example, a voice memo recorded on a phone in a kitchen while the fridge is running. Hum and hiss go away in one or two passes, and the fridge is softened to a level that disappears under speech. The echo from the tiled wall remains and needs its own step. That result is normal: a home recording never reaches perfect silence.

How to check that cleaning has not damaged the voice

Check in three passes, because different faults show up at different volumes and on different playback systems.

First, listen quietly. Play the result at a level where speech is barely audible and try to make out every word. If the consonants "s", "t" and "f" vanish at that level, the cleaning has taken away too much of the top end.

Second, listen on a phone or a laptop. A small speaker reveals the bubbling and metallic edge that headphones sometimes hide. If long vowels sound pulsed, that is a tonal leftover of the noise, and the noise reduction should be eased off.

Third, compare with the original. Switch between the two versions on the same place: breath, pauses, the start and end of a phrase. A good cleanup makes the background less noticeable but does not change how the voice sounds overall. If the voice has changed, sounding further away or losing its life, the processing went too far.

When cleaning will not help

Noise reduction works with what has already been recorded. There are situations where it will not gain you anything, and it is better to recognise them early:

  • The voice is barely audible above the noise. When the noise is close to the speech in level, lowering the noise takes the voice down with it.
  • The recording is overloaded and clipped. Clipped peaks cannot be restored by noise reduction, and the processing only adds artefacts.
  • Speech is covered by another voice or by music in the same frequency range. Separating them is only partly possible, and that is a job for stem separation, not noise reduction.
  • Echo in a small room with bare walls. Reverberation like that often makes the recording unintelligible, and it is simpler to record the passage again.

In such cases it is cheaper to record again, or at least that passage, than to fight the file for a long time.

How to do it in Magic Master

The restoration page is built around three kinds of problems: 50 or 60 Hz mains hum, tape hiss and vinyl clicks. For voice recordings made at a computer, hum and hiss are the most common, so start there.

Base mastering can be run for free within the quota: three times per calendar month for guests, once per day for signed-in accounts without tokens, and three times in the first 24 hours after sign-up. PRO modules cost tokens: one token buys one full master with all PRO modules. If the cleaned file also needs loudness and tone finishing, mastering in the app does that within the same base quota.

How to record cleaner next time

A clean recording is always easier than a repaired one, so a few minutes of preparation pay off immediately:

  • Keep the microphone close to the voice. The closer the source, the fewer room reflections and the higher the voice sits above the noise.
  • Switch off the fridge, fans and air conditioner. Test the room: record ten seconds of silence and listen at full volume.
  • Use a cardioid microphone and turn it so that the noise sources end up behind it.
  • Soften the room: a rug, curtains and soft furniture reduce reflections. Bare walls produce echo that no tool removes cleanly.
  • Move the microphone away from the computer case and any vents: fan noise reaches the mic through the air and through the desk.
  • Keep the microphone slightly off-axis from the lips or use a pop filter: fewer clicks and plosive pops.
  • Record uncompressed, in WAV or FLAC. A low-bitrate file has already lost detail, and noise reduction works worse on what is left.
  • Record loud but not clipping: peaks should never hit the ceiling. A quiet file has to be raised later, and the noise rises with it.

Common mistakes

  • Pushing noise reduction to the maximum: the voice turns "underwater", and noise is still audible on the consonants.
  • Cleaning after normalisation and compression: the noise has already been raised and mixed into the voice.
  • Trying to remove echo with a regular noise reducer: almost no effect, and real damage to the voice.
  • Total silence in the pauses: holes appear, and listeners notice the sudden drop in room tone.
  • Cleaning without comparing to the original: it is easy to keep a version where the consonants are already gone.

A recorded voice can almost always be made cleaner, but rarely perfect. The less noise gets into the file, the simpler the cleanup and the more natural the voice in the end.

Questions and answers

How do I remove noise from audio online?

First identify the kind of noise: steady hum and hiss clean up well, while clicks and other voices do not. Work in small steps and listen on headphones after each one.

Can I remove background noise from a recording without losing quality?

Not entirely: any noise reduction changes the signal too. The goal is to remove the interference and keep the voice natural, not to reach absolute silence.

How do I remove echo from a microphone recording?

Echo is reflections from walls and ceiling, and ordinary noise reduction will not remove it. It needs a separate de-reverb step: see remove reverb from audio online.

How do I remove noise from audio?

Start by finding a passage of silence, because it shows what the noise actually is. Mains hum, hiss and room tone are removed using a sample of that noise, while clicks and echo need their own tools.

Try Echoteka

Upload a track — the same mastering engine, prices in euros.

Open Echoteka →AI mastering onlineLUFS analyzer