Audacity Noise Reduction: Remove Background Noise and Know When to Stop
Most guides hand you settings. This one teaches you to hear when the reduction has gone one step too far.
Audacity's Noise Reduction effect works in two moves: you show it a piece of pure noise, then you tell it how hard to push. Most guides hand you numbers for the second move and rush the first, which is backwards, because the profile decides almost everything that follows. This is a walkthrough for a file you have already recorded and can open right now, not advice for anything happening live. We will cover where the effect sits in the menu, what each of the three sliders actually changes, and the sound of having gone too far, since the failure mode here is not an obviously broken file. It is a quietly ruined one that survives a quick listen.
The noise profile is not a step, it is the whole job
Audacity needs a noise profile because it has no idea what your room sounds like until you show it. You select a stretch containing nothing but the noise, run Get Noise Profile, and the effect builds a frequency-by-frequency picture of what to subtract. Hand it a bad sample and every setting afterwards is guesswork applied to a wrong answer. Nearly every complaint about Audacity ruining a recording traces back to this one selection, not to the sliders people blame.
- Two seconds of noise alone beats half a second. The effect gets a steadier picture of what it is meant to remove.
- The sample must come from this file, this room, this session. Noise fingerprints do not travel between takes or days.
- No breaths, no chair creak, no distant word. Anything inside the sample is classified as noise and stripped everywhere.
- Room tone at the head of the recording is usually the best candidate you have.
Here is what a contaminated profile does. Include the tail of a word and the effect learns part of your voice as noise, then subtracts that shape from every syllable in the file. The result is not obviously wrong. It is thin, slightly hollow, missing a band you cannot name, and it will bother you for a week without your working out why. Include a breath and consonants go soft. This is also why a file recorded across two sessions wants two profiles, applied to the two halves separately.
The three sliders, and what each one costs you
Reopen the effect after capturing the profile and you get three controls. Noise Reduction, in decibels, is how much to subtract; it is the one everybody drags to the right. Sensitivity governs how readily the effect decides a given sound qualifies as noise, so raising it catches more of the quiet detritus and also more of your consonants. Frequency Smoothing, in bands, spreads the reduction across neighboring frequencies, which softens the isolated tinkling artifacts at the cost of some depth. The three interact, so change one at a time and preview after each.
Start low, and let a second pass do the work
We set Noise Reduction low first, keep Sensitivity modest, and leave Frequency Smoothing at a few bands. Preview, listen to the gaps between words rather than to the words themselves, then decide. A second gentle pass over an already-treated file sounds better than one aggressive pass, because each pass has less to remove and pushes the algorithm less far into the invention that produces artifacts. Big single moves are how good takes become unusable.
How to remove background noise in Audacity, step by step
Audacity is free, open source, and runs entirely offline, so nothing here uploads anywhere. Save a copy of your project before you begin, because the effect writes onto the audio and enough undo steps will eventually run out. Work with headphones on. Laptop speakers hide precisely the artifacts you are trying to catch, and the whole point of this method is catching them while you can still change your mind.
- Select two seconds of silence with the mouse, dragging inside the waveform.
- Open Effect, then Noise Removal and Repair, then Noise Reduction.
- Click Get Noise Profile. The dialog closes on its own; that is expected, not a crash.
- Select the whole track, reopen the same effect, set the sliders low, and press Preview.
- Toggle Preview on and off against the untreated audio before you commit with OK.
If the file has no silent passage at all, this method has nothing to work with and you need a different route: an AI enhancer such as Adobe Podcast Enhance, or one of the browser-based online tools, neither of which asks for a profile. A steady electrical buzz is a separate problem with a narrower fix, and hum and hiss are worth handling on their own terms. Faint archive material behaves differently again, so old recordings get a gentler treatment. If your audio is locked inside a video file, extract it before you start and see our notes on fixing video audio for the reassembly.
Hearing when you have gone too far
The watery, underwater sound
This is the classic overshoot. Small tinkling tones appear in the gaps between words, wandering in pitch like a wind chime heard through a bathtub. They are fragments of noise the effect could not fully remove, left ringing on their own. Lower the reduction amount or add a band or two of frequency smoothing until they stop.
Consonants that lost their edges
S and T sounds live high and quiet, exactly where the effect goes looking for hiss. Push too hard and speech turns mushy, lisping on every plural. Read a sentence with several plurals and listen to those alone.
Silence that pumps
Between phrases the file goes completely dead, then the room whooshes back in the moment somebody speaks. Real rooms never do that. A faint constant floor is the sound of a recording that was cleaned by somebody who knew when to stop, and comparing tools on this one behavior is more useful than comparing their feature lists.
Frequently asked questions
Where is noise reduction in Audacity?
Under the Effect menu, in the Noise Removal and Repair submenu, listed as Noise Reduction. Older releases put it directly in the Effect list, so if your menu looks flatter than the description you are on an earlier version. The effect is greyed out until you have selected some audio in a track.
Why is Get Noise Profile greyed out?
Because nothing is selected, or the selection is too short for the effect to analyze. Drag across a stretch of the waveform, ideally a couple of seconds of noise with no speech, and reopen the effect. If it stays unavailable, check that you clicked inside the audio track itself rather than the label or the panel beside it.
How much noise reduction should I apply?
Less than feels satisfying. Start low, preview, and add only if the noise still distracts you from the words. The target is not silence between phrases, it is a quiet, even floor you stop noticing. If the result sounds impressively clean on first listen, check it again on headphones an hour later.
Can Audacity remove people talking in the background?
Not really. The effect subtracts a fixed frequency picture, and speech is neither fixed nor separable from your own voice by frequency alone. Profiling a stretch of the other person and applying it will gouge holes in your speech too. Background chatter is a case for an AI enhancer, or for recording the take again.
Does Audacity noise reduction work on music?
Cautiously. Music occupies far more of the spectrum than speech, so the effect has less room to work without touching the instruments. Reverb tails and cymbals suffer first, turning grainy. Keep the reduction amount much lower than you would for a voice, and expect to accept some remaining noise as the better trade.