Noise Reduction
Noise reduction models the steady layer under the voice — hiss, hum, room tone — and subtracts it band by band.
- Free to start
- No install
- Built for hiss, hum and room tone
What is noise reduction, and who is it for?
Noise reduction is the process of separating a sound you want — nearly always a voice — from the constant background it happened to be recorded against. On the spectrogram bench above you can see what that means: the bright bands are speech, the dim wash across every band is the noise, and after the pass the wash falls away while the speech is untouched.
It is for people who record in the real world rather than in a booth: podcasters, interviewers, teachers, journalists working off a phone, anyone narrating over a laptop microphone, archivists digitising old tape. If you have ever finished a good take and then noticed the fridge, this is for that.
Pixazo runs it inside Pixazo, alongside the rest of the audio work in the Pixazo AI tools directory — so a cleaned take can go straight on to a voiceover project or sit next to sound you built with the AI audio generator.
What it does not do: it is not a repair shop for one-off sounds. A door slam, a cough or a siren passing the window has no steady pattern to model, and no denoiser removes those cleanly.
On the bench — at a glance
Spec| Input | Your recorded audio — interview, voiceover, narration, archive transfer |
| Targets | Steady, continuous noise: hiss, mains hum, room tone, air-conditioning, fan and drive noise |
| Strongest on | A noise that never changes across the take |
| Weakest on | Bangs, slams, sirens, overlapping speech, clipped or near-silent audio |
| Runs in | Browser — Pixazo, desktop and mobile |
| Output | The same recording with the background layer subtracted |
| Price | Free to start · scales with your Pixazo plan |
Why can steady noise be subtracted — and one-off noise not?
This is the whole engineering idea behind the pass, and it is also the honest boundary of it. Steady noise has a fingerprint. A one-off event does not.
A steady noise has a fingerprint
Hiss, hum and room tone hold roughly the same shape in every band for the whole recording. Because that shape is predictable, the engine measures it once and takes exactly that much out everywhere — which is what drops the noise floor on the bench.
A one-off event has none
A slam, a chair scrape, a cough or a siren happens once, with a shape nothing else in the take shares. There is no profile to build, so a denoiser either misses it or claws a hole in the audio around it. Cut those moments in an editor.
Steady hiss, hum and room tone come out; the voice stays.
Remove Background Noise freeHow do you run noise reduction on a recording?
Four moves on the bench: open the studio, upload the take, run the pass, and compare before you export.
Open the studio
Open Pixazo in your browser on desktop or mobile. Nothing to install, and no plugin to load into an editor first.
Upload the recording
Drop in the interview, voiceover or tutorial audio you want cleaned. Use the best version you have — the original file, not a re-compressed copy.
Run the denoise pass
The engine reads the constant background under your voice, builds a profile of it band by band, and subtracts that profile across the whole take.
Compare, then export
Listen to the cleaned take against the raw one. If the voice sounds thin, back the strength off and run it again, then download the result.
The four takes this actually fixes
Every one of these is the same problem wearing a different hat: a constant sound that was in the room, or in the circuit, the whole time.
An interview recorded beside an air-con unit
The unit never stops, so its rumble and fan hiss sit under every answer at the same level. One profile covers the whole conversation.
A voiceover in an untreated room
Bare walls give you a low, continuous tone plus reflections that make a read sound boxy. A pass lifts the tone away; the reverb stays, because that is your own voice arriving late.
A tutorial narrated on a laptop microphone
Built-in microphones sit inches from the fan, the drive and the desk. A pass here is often the difference between a usable screen recording and one people click away from.
Tape hiss on an old recording
Tape adds a broad, unchanging hiss across the top of the spectrum — textbook material for a noise remover, and why archive transfers give the most dramatic before-and-after.
Where does the noise come from?
Three of the usual suspects — each one a steady source, which is exactly why a profile works on it.
The honest limits of noise reduction
Read this before you plan a shoot around it. These are not settings problems; they are the boundary of what subtracting a profile can do.
- One-off sounds stay. Bangs, slams, sirens, dropped cutlery and coughs have no repeating fingerprint. Cut them out on the timeline instead.
- Push it hard and the voice suffers. Speech shares frequencies with the noise, so aggressive settings make a read sound thin, watery or hollow. Restraint beats strength every time.
- It cannot unmix two voices. Two people talking over each other are two wanted signals, not signal plus background.
- Clipped or near-silent audio stays broken. Distortion has erased the waveform and a whisper-level take is already buried in the floor.
- Reverb is not noise. A boxy room adds late copies of your own voice; that is a different problem, and this pass will not fix it.
How much reduction should you apply?
The one judgement call on the bench — and the answer is almost always less than you think you want.
| Setting | What you hear | What it costs |
|---|---|---|
| Rail 01Light | The wash sits back, the voice is identical to the raw take. | Some noise remains in the gaps — usually inaudible under speech. |
| Rail 02Moderate | Background almost gone, the read still sounds like a person in a room. | A faint softening of breaths and sibilance. This is the setting to aim for. |
| Rail 03Heavy | Silence between words is dead quiet. | Thin, watery, underwater artefacts on the voice. Almost never worth it. |
Judge it on the voice, not on the silence. The gaps between words are the easiest place for a pass to look impressive and the worst place to decide from. Play a full sentence, on headphones, and ask whether the read still sounds like a person in a room.
Noise reduction — frequently asked questions
The questions people ask before they upload a take.
Is noise reduction free to try?
You can start free with no credit card in Pixazo. Longer files and heavier use follow whichever Pixazo plan you are on.
What kind of noise does it actually remove?
Steady, continuous noise: tape and preamp hiss, mains hum, air-conditioning, fan and computer noise, and the general room tone of an untreated space.
Can it remove a door slam or a passing siren?
No. A one-off bang, a slammed door, a siren going past or a chair scrape has no repeating pattern to model, so it cannot be subtracted cleanly. Edit those moments out instead.
Will it make my voice sound thin or watery?
It can if you push it too hard. Speech shares frequencies with the noise being removed, so heavy settings take some of the voice with them. Restrained settings sound far better.
Can it separate two people talking over each other?
No. Overlapping speech is two wanted signals, not signal plus background, and this tool cannot pull them apart. Record each speaker on their own track if you can.
Does it work on the audio from a video?
It works on audio. If your material is a video, take the audio track out, clean it, and lay the cleaned version back against the picture in your editor.
Is this the same as the noise cancellation in headphones?
No. Headphone systems fight sound in the air before it reaches your ears. This works after the fact, on a file, and it can only touch what the microphone already committed to tape.
Can it rescue a recording that clipped or was almost silent?
Not really. Clipping destroys the waveform and a near-silent take has the voice buried in the noise floor, so there is little clean signal left to recover.
Do I need any audio experience to use it?
No. Upload, run the pass, listen to both versions and export. The one judgement call is how much reduction to apply, and less is usually the right answer.
Drop the noise floor, keep the voice.
Bring the take you already have and let the studio lift the hiss, hum and room tone out from under it. Free to start, nothing to install, both versions side by side before you commit.
Remove Background Noise- Free to start
- Steady noise, honestly handled
- Compare before you export
