How to Reduce Noise in Audacity
Reduce noise in Audacity with a clean noise profile. Hear what the filter removes, fix weak settings, and export speech without chasing complete silence.

On this page
To reduce noise in Audacity, capture a sample of background sound, select the recording you want to clean, and apply Noise Reduction. You'll need an audio file with some noise-only space. Start with a copy so you can compare the voice before and after processing.
Reduce noise in Audacity by listening to what gets removed
Audacity can play the audio it plans to discard. The current manual calls this output Noise only. If that contains pieces of the words you wanted to keep, the filter is taking too much. This is a better check than judging the size of a waveform after processing.
The Audacity Noise Reduction manual describes both passes. Older versions and tutorials use Residue for the checking output. This guide uses the manual checked October 9, 2026; the Audacity project page has the directory details.
My pick is a light cleanup that leaves the speaker recognizable. A faint fan between sentences can be less distracting than a voice that breaks into metallic fragments.
1. Open a copy and listen to the noise
Use File > Open to load the recording. Audacity's current File manual confirms that opening an audio file creates a project containing that audio. Give your working copy a name that separates it from the original recording.
Listen to a pause, the start of a sentence, and a quiet word ending. Decide what you're actually hearing: steady hiss, a low hum, or a single interruption. Noise Reduction targets steady background sound. A door slam is a different repair job.
Keep the original available outside the project. For example, place voice-original.wav and the project in separate folders. That lets you reopen the untouched recording even after saving edits and closing Audacity.
2. Select a stretch containing only background sound
Drag across a pause with no speech, music, or wanted sound. A few seconds of the same fan or hiss heard beneath the voice gives the effect a sample to learn from.
Don't use a breath, a faint word, or the tail of a musical note as the profile. Those sounds would teach the filter that part of your recording is unwanted. Zoom in when the gap is short so the selection doesn't include the next syllable.
If the room sound changes halfway through, treat the two parts separately. A profile from a quiet opening won't describe a refrigerator that starts humming during the final paragraph. The older Noise Reduction manual documents separate profiles for different noise conditions.
3. Capture the noise profile
Open Effect > Noise removal and repair > Noise reduction, then choose Get noise profile. The dialog closes after learning from the selection. That first pass creates the profile; it doesn't clean the spoken section yet.
Now select the audio you want processed. This second selection should include the speech that needs cleanup, not only the noise sample. A narrow selection left on the original pause will produce a cleaned pause while the rest of the recording stays untouched.
Keep that selection visible when reopening the effect. If you only want to fix one interview answer, you don't need to process a second speaker whose recording is already quiet.
4. Adjust the three controls on a short passage
Reopen Noise reduction. The manual lists Noise reduction, Sensitivity, and Frequency smoothing. Their defaults are 6 dB, 6, and 6 bands in the current documentation. Treat those as a starting position, not a recipe for every microphone.
Noise reduction controls how much the learned sound is lowered. Sensitivity controls how readily sound is classified as noise. Frequency smoothing spreads the treatment across neighboring frequency bands. Move one control at a time so you can hear what changed.
Use a passage with both loud and soft speech. Listen for word endings and the spaces between them. A setting that sounds acceptable on one shouted phrase can erase softer consonants elsewhere in the same take.
5. Check Noise only, then return to the cleaned output
Set Output to Noise only and apply the effect to the short passage. Listen to that result. Undo the operation, reduce the strength if wanted speech was present, and reopen the effect.
Choose Audio with noise removed for the finished treatment. After applying it, compare against the unprocessed recording at a similar listening volume. Louder playback can make a worse recording sound more impressive, so don't use volume alone as the decision.
There is no prize for a perfectly flat gap between sentences. Judge whether the listener can follow the words without the background pulling attention away from them.
6. Export a separate audio file
Use File > Export audio for the finished recording. The File manual separates this from saving the editable project. Select the full project or the intended audio selection in the export dialog, then save the result under a new filename.
Open that file in another player, such as VLC, and hear the beginning and ending. If the cleaned voice belongs in a video, import it into Shotcut beside the picture and compare the sync before replacing the camera sound.
Known problems and fixes
In an April 2026 forum thread, a user on Audacity 3.7.7 found that cleanup had stopped changing the noise floor. Sensitivity had fallen to 0.01; returning it to 6 restored the effect. If processing seems inactive, read the stored value before taking another profile.
That thread also explains why choosing the discarded-audio output can leave only the unwanted sound in the track. Undo it before exporting, then choose the cleaned output. Closing and reopening the project won't turn the discarded audio back into the speaker's voice.
| Problem | Likely cause | Fix |
|---|---|---|
| Effect appears inactive | Very low Sensitivity or the wrong selected region | Read Sensitivity and select the spoken passage |
| Voice has watery edges | Treatment is too strong | Undo and lower reduction or sensitivity |
| Only noise remains | Checking output was applied | Undo, then choose Audio with noise removed |
| One loud interruption survives | Sound isn't steady background noise | Edit that interruption separately |
Variations for different recordings
For narration over music, clean the voice recording before mixing in the music. That gives you a noise sample without the music teaching the profile unwanted frequencies.
For an interview with separate microphones, process each track using its own room sample. The quieter microphone doesn't need to inherit the settings of the noisy one.
For a video with only basic speech cleanup, VEED offers a browser route. Creative & Design and the video editing category contain related finishing tools.
Completion checklist
- The noise profile excludes speech and music.
- The intended spoken region received the effect.
- Quiet consonants survive the treatment.
- The checking output was undone before the final pass.
- The exported file contains the complete intended recording.
- The original audio and editable project are still available.
FAQ
Can I remove traffic or a slammed door this way?
Noise Reduction is designed for steady noise, so a sudden door slam needs a separate edit. A changing traffic recording can also resist one profile. Find the location of the interruption and decide whether you can remove that section, replace the phrase with another take, or use a more specific repair tool.
Does saving the project make a playable WAV?
No. Saving keeps the editable Audacity project; exporting creates an ordinary audio file. The current File menu calls the latter operation Export audio. Keep the project when you may need to change the edit later, and send the exported file to the person or application that needs to play the finished recording.
Next steps and sources
Add the cleaned recording to your edit and check a lip movement or visible action against the sound. Audacity's current Noise Reduction and File manuals, the older Noise Reduction manual, and the linked April forum discussion support these steps. Official controls checked October 9, 2026.
Related articles
Directory
Listings in this article
-
Audacity
Audacity is a free, open-source audio editor and recorder. It can cut, trim, split, reverse and mix audio, apply effects such as reverb, compressor and distortion, reduce noise and remove vocals. It runs on Windows, macOS and Linux, supports major audio formats and third-party plugins, and includes AI transcription.