How to Remove Background Noise in Audacity

On this page

1. Open the recording to remove background noise in Audacity #

Open Audacity and load the audio with File → Open, or drag the file into the editor. The waveform should appear in a track and play when you press the transport's play control.

For this worked example, use your own one-minute announcement. Leave about two seconds of room sound at the beginning, followed by: “The workshop starts at ten. Bring your laptop and the project files.” Record the example in your room so you can compare the untreated voice with each adjustment.

Save an editable working copy and keep the source recording separate. That gives you a way to compare the voice before and after processing. A cleanup pass should improve the recording without forcing you to accept damage to the words.

Listen once before opening an effect. A steady fan is a reasonable target. A barking dog or a door slam requires a different repair decision because it isn't a constant noise bed.

2. Select noise without speech #

Drag across the opening room sound in the waveform. Leave out breaths, handling noise and the first syllable. The selection should contain the fan by itself.

The current Noise Reduction manual explains that the effect first learns a noise spectrum, then filters the selected recording. Any voice in the profile can become something the filter tries to remove.

Replay the selected region. If you hear a word or a breath that you want to preserve, shorten the selection or find another quiet region. A good sample matters more than a high reduction value.

3. Create the noise profile #

Choose Effect → Noise removal and repair → Noise reduction in the current Audacity 4 interface. Press Get noise profile. The dialog closes after learning the selection.

That step hasn't cleaned the track yet. It has stored the noise information for the second pass. Select the audio you want to process next. Ctrl+A selects all audio on Windows and Linux; use Command+A on macOS.

If the room sound changes halfway through the recording, process sections with their own profiles. For example, a fan switched on during an interview is a different noise environment from the first quiet half.

4. Start with a modest reduction #

Open Noise reduction again. The current dialog has Noise reduction, Sensitivity, Frequency smoothing and Output controls.

I'd begin with the current defaults, then adjust by listening. The manual lists 6 dB for Noise reduction, 6 for Sensitivity and 6 bands for Frequency smoothing. Listen for damage to consonants and breaths before applying the effect to the whole recording.

Choose Audio with noise removed under Output. The aim is less fan hiss with clear consonants. It doesn't need to sound like a silent room between every word.

Increase one control at a time. If the voice becomes watery, undo the change and use less reduction. Large simultaneous changes make it harder to identify which setting damaged the sound.

5. Listen to what the effect would remove #

Choose Noise only under Output and apply the effect. Listen to a sentence with softer consonants and a pause. You should mainly hear the fan or hiss. Recognizable words mean the processing is treating wanted audio as noise.

Use Edit → Undo to restore the recording after this check. Reopen Noise reduction, lower the settings if words were removed, and choose Audio with noise removed under Output. The current manual describes using Noise only followed by Undo to inspect what the filter removes.

Confirm the output choice before applying the cleaned pass. Leaving Noise only selected produces isolated noise instead of the announcement.

The Audacity noise-repair documentation also describes artifacts from a poor profile or unsuitable sensitivity. A cleaner noise sample can solve more than another increase in strength.

6. Apply the effect and compare the whole phrase #

Accept the effect, then play the voice and pauses. Use Edit → Undo if it sounds worse. Return to the same region and change one setting before applying again.

Compare at a similar listening volume. A quieter file can seem cleaner simply because everything is quieter. Listen for the start of “workshop,” the ending of “starts” and the softer phrase “project files.” Those details should remain easy to understand.

Check the end of the recording too. A noise profile chosen at the start may work poorly after the speaker moves closer to the microphone or another appliance begins running.

7. Export a copy and check it outside the project #

Choose File → Export audio to create a separate audio file. Under Type, choose Export full project audio so a remaining selection doesn't limit the output. Choose WAV when the next step is more editing, or a sharing format for the final destination. The current File-menu documentation covers these export controls.

Open the exported file in a separate player such as VLC. Confirm that the whole announcement is present, the noise-only output wasn't exported by mistake, and the result has the expected duration.

If this voice belongs in a video, bring the cleaned copy into the video project and check it against the picture. Our video-conversion guide covers preparing the final sharing copy after editing.

Reported problems and practical repairs #

A September 2026 audio-restoration discussion described a muffled voice after aggressive reduction. Back off the strength and repeat the Noise only check, followed by Undo. Removing the last trace of hiss is a poor trade if the sentence becomes harder to understand.

In a November 2025 Audacity Linux discussion, users worked through unexpected reduction behavior after an upgrade. The repair path starts with a fresh noise profile and the correct selected region. If behavior changed with an update, record the application version and reproduce the problem on a short copy before changing the whole project.

People looking for an AI alternative also asked about OpenVINO in the Audacity 4 feedback discussion. The official OpenVINO page says those plugins aren't yet compatible with the latest Audacity. Installing an incompatible bundle won't improve this two-pass workflow.

Troubleshooting #

When the recording needs another approach #

Treat a hum separately

A narrow electrical hum and broad fan hiss aren't the same target. The current effect menu also includes Notch Filter for a specific tone. Identify the sound before adding another processing pass; an unnecessarily wide notch can remove part of the voice.

Process the quiet and noisy sections differently

For an interview spanning two rooms, make a profile for each room and work on selected sections. Compare the transition between them. The listener shouldn't hear a sudden change in voice tone simply because a new cleanup setting begins there.

Choose a simpler editor for basic cuts

If you only need to remove a pause, ocenaudio may be quicker. AudioMass offers a browser route for a short edit. Neither decision requires applying noise reduction when the recording is already clear.

Completion checklist #

• The profile contains only unwanted steady noise.
• The Noise only check doesn't contain recognizable speech, and Undo restores the recording afterward.
• Soft words and sentence endings remain clear.
• The final output uses Audio with noise removed.
• The exported copy plays in another application from beginning to end.
• The editable project and source recording are still available.

Next steps #

Compare Audacity and ocenaudio for layered work and single-file edits, or explore the wider free audio-editor shortlist.

For a layered song or collaborative recording, compare the roles of GarageBand and BandLab. For moving your finished recording between devices, the file-transfer shortlist covers local and online handoffs.