Skip to content
Gleam Audioby purplecandy

Search the docs

Download Gleam Audio

Free for personal use, with every tool, up to 20 files a week. You only need a license for work.

Version 2.2.0 · Nov 8, 2026

Windows

macOS

Linux

Run this in a terminal to install it for your user:

curl -fsSL https://static.gleamaudio.com/gleam-audio/releases/latest/download/install.sh | bash

Or download a file below and install it by hand.

Windows packages are not code-signed yet, so SmartScreen will ask you to confirm: More info, then Run anyway. Every file has a checksum in SHA256SUMS.

Gleam AudiohomeGuide

Noise Reduction With Audacity: Settings That Work

Audacity's Noise Reduction works in two passes: profile the noise, then filter the recording. The exact controls, the defaults, and what to do when it fails.

By 5 min read


Audacity’s Noise Reduction is the free route most people start with, and it works well once you understand the one thing it asks of you: a few seconds of the noise on its own, before any voice.

That sample is the whole trick. The effect learns the shape of the noise from it and subtracts that shape from the recording. Give it a clean sample and three sliders finish the job. Give it a sample with a voice in it and the filter takes the voice out too, which is where the metallic sound people complain about comes from.

The steps below follow the manual for Audacity 4.0.1, the current release on 1 November 2026. Audacity is free and open source, and runs on Windows, macOS and Linux.

What you need

  • Audacity 4.0.1, from audacityteam.org.
  • A recording that has a few seconds where only the noise plays: the fan, the hum, the room, no talking. Most takes have one at the start.
  • Headphones. Speakers make it hard to hear what the filter removed.

1. Select a stretch of noise alone

Open the recording and find the quietest stretch before the first word. Drag across one to three seconds of it, with the pointer. Short is fine; what matters is that nothing but noise is inside the selection.

If the take has no such stretch, record ten seconds of silence in the same room at the same settings and use that instead. The room tone has to come from the same session, because the noise changes with the mic, the gain and the room.

2. Press Get noise profile

With the noise selected, open Effect > Noise removal and repair > Noise reduction. The dialog is laid out as two steps. The first one is the button: press Get noise profile.

Audacity reads the selection and remembers its spectrum. Nothing has been changed in the recording yet. The profile stays in memory for the session, so the second step can be run on the whole file after it.

3. Select the audio to clean and set the sliders

Select all of the recording, or just the part you want cleaned, then open the same effect again. Now the dialog does the work. Three sliders control it:

  • Noise reduction is how much the noise is turned down, in dB. Range 0 to 48, default 6.
  • Sensitivity is how readily a sound is judged to be noise rather than signal. Range 0.01 to 24, default 6.
  • Frequency smoothing is how far the reduction is spread across neighbouring frequency bands. Range 0 to 12 bands, default 6.

Leave all three at the defaults the first time and listen. The defaults are deliberately modest, and a modest reduction that leaves a little noise behind usually sounds better than a total kill.

4. Listen to what is being removed

At the bottom of the dialog, Output offers two choices: Audio with noise removed and Noise only.

Noise only is the setting to reach for when the result sounds wrong. Apply it, play the file, and listen for the voice inside it. If you can hear words in the noise track, the settings are too aggressive for that recording: undo, lower Noise reduction, and run it again. This check takes a minute and saves the take.

5. Apply, then listen in the gaps

Press the button that applies the effect, and listen to the file, not to the dialog. The place to judge the result is the space between words, where the noise was loudest. It should be quiet without sounding gated or choppy.

Then listen to a sentence. The consonants should still be there. A voice that has lost its edges, or that warbles on long vowels, means the reduction went too far.

When it won’t help

  • One-off sounds. The manual is explicit: the effect suits constant background noise and will not remove a door slam. Use Repair for a single click or thump, or cut the sound out in the waveform.
  • Noise that comes and goes. One profile describes one spectrum. Traffic that swells, a fan that changes speed or a fridge that cycles will be partly left behind, and a profile set to catch the loud moments will eat into the quiet ones.
  • A profile from another room or another mic. The spectrum it learned does not match the take. Profile again from the file you are cleaning.
  • Nothing selected. The effect expects a profile first and does not fall back to the whole project when nothing is selected, so an empty selection gives you nothing to apply.
  • A recording where the noise sits on top of the voice. Hisss over a whole take is handled well. A second person talking, music under the speech, or a TV in the room is not noise to this effect, and no setting will separate it.

When you have a folder of files

Audacity’s effect is one file at a time in the editor, with a profile you select by hand. If the job is a batch, a week of episodes or a folder of interviews, the setup work repeats for every file.

Gleam Audio takes the other approach. Its Remove noise tool uses a model trained on human speech, so there is no sample to profile and no spectrum to match: pick the files, set Strength, and run. It works on the whole list at once, on your own computer with nothing uploaded, and it never changes the originals. The comparison with Audacity’s effect, control by control, is in Gleam Audio vs Audacity, and the two-way comparison with Adobe’s browser tool is in Audacity vs Adobe Podcast.

To hear the same model on one short file first, the free noise remover runs in the browser with no install.

Takeaways

  • Noise Reduction needs a stretch of noise alone. The profile is the part that decides the result.
  • The defaults (6 dB, Sensitivity 6, 6 bands) are a good first pass. Raise Noise reduction slowly.
  • Use Output > Noise only to hear what is being taken out. Words in that track mean the settings are too high.
  • Steady noise is what this effect is for. Door slams, coughs and music are not.
  • For a folder of files, a tool that needs no profile per file saves the setup time.

Questions

What are the best settings for Noise Reduction in Audacity?

Start at the defaults: Noise reduction 6 dB, Sensitivity 6 and Frequency smoothing 6 bands. Raise Noise reduction only until the gaps between words are quiet, and stop before the voice starts to sound hollow.

Why does Audacity's Noise Reduction make my voice sound metallic?

The filter is working too hard. A high Noise reduction value removes part of the voice along with the noise, and what is left sounds watery or metallic. Lower the value, and check that the profile came from noise alone.

Do I need to select noise before using Noise Reduction?

Yes. The effect expects you to profile a small stretch that contains only the noise, then apply it to the audio you want to clean. With nothing selected it does not fall back to the whole project.

Does Audacity remove a door slam or a cough?

No. Noise Reduction is built for steady noise such as hiss, hum or a fan. One-off sounds need the Repair effect, or a cut around the sound in the waveform.

Comments


Gleam Audio Pro

Hey, we're working on something new. We're calling it Gleam Audio Pro: a multi-track editor for podcasts, lectures and narration. It's our take on an editor for busy creators, so you spend your time on the show, not the knobs.

Gleam Audio isn't going anywhere. It will keep getting updates and stay focused on one thing: studio-quality sound, without the studio.

Pro is coming to Mac and Windows. Join the waitlist →

By signing up you agree to the privacy policy.