Saturalabs / Specific distractions
Remove background noise from recorded audio.
Try separating a distracting sound from a recording while keeping the audio you need. Describe one audible source, compare both outputs, and listen closely to the moments where sounds overlap. AI processing requires an account and credits.
Start with your recordingThe track you’re looking for
The remaining recording after the requested sound is separated. Check the wanted speech or music against the original.
Name one audible source. You can edit this prompt before running the separation.
You can open the editor before signing in. Running isolation requires an account and enough credits. Downloads use credits. View pricing.
A focused approach
Choose what the recording needs to retain
A recording can contain wanted room sound, music, speech and an unwanted fan or other distraction. If you want to retain more of the scene, name the distracting source—for example, ‘The background fan’—and review Background audio, the remaining mix. Replace the example with what you actually hear.
If you only need spoken words, a second option is to describe the foreground speaker and review Isolated sound. That can also reduce ambience, music and other wanted sounds. These are different candidates, not settings that guarantee complete noise removal.
This is text-guided source separation, not a noise-profile filter or live microphone suppression. A steady hiss or electrical hum may respond differently to a dedicated noise-reduction or notch filter; compare methods on a copy of the original.
From mix to source
How to remove background noise from audio
Choose a representative recording
Upload a local MP3, WAV, MP4 or MOV up to 50 MiB (52.43 MB). Keep the original, and include a quiet phrase and a moment when the distraction overlaps wanted sound.
Name one audible source
Describe the distraction, such as ‘The background fan’ or ‘The dog barking.’ The fan prompt is a starting example, not a preset; replace it if your recording contains a different sound.
Listen to the remaining mix
For an unwanted-sound prompt, review Background audio. Isolated sound is the requested sound. If you instead request the speaking voice, use Isolated sound for the voice candidate.
Compare and choose
Listen at a similar level to the original. Check words, music and room tone during the loudest overlap. AI processing requires an account and sufficient credits; the results screen shows the WAV download cost.
Be specific about the sound
A prompt for the result you want
Describe the source, then choose the output that matches your goal. These are starting points to try with your own recording.
| Your goal | Try this prompt | Keep this track |
|---|---|---|
| Reduce a fan while retaining the recordingCheck that the fan is audible and that wanted speech or music survives. | The background fan | Background audio |
| Reduce an intermittent barkListen to each bark that overlaps the wanted sound. | The dog barking | Background audio |
| Focus on the speakerA different goal: music and ambience may be reduced too. | The foreground person speaking | Isolated sound |
For a custom target, prepare a sound description and choose the output to inspect. The helper opens in a new tab so you can keep your selected recording here.
Made for a real next step
Put the separated audio to work
Clean a recorded explanation
Try reducing a fan or intermittent event behind a tutorial or voiceover, then check quiet word endings and breaths in the remainder.
Keep more of a scene
When footsteps or room tone matter along with dialogue, compare a named-noise remainder with a voice-only candidate. Listen for missing scene details in both.
Compare a noisy music passage
If one distinct sound competes with an instrument, test that sound as the target. Listen for changed instrument attacks and decay; source separation is not a mastering or restoration guarantee.
Listen before you commit
Three checks worth making
Compare a quiet passage and a busy passage with the original, at a similar listening level.
Quiet speech
Compare consonants, breaths and word endings. A quieter background is not useful if the message becomes harder to understand.
Overlap and remnants
Listen where the unwanted sound is loudest, including its start and tail. It may remain or take wanted sound with it.
Context and continuity
Check music, room tone and transitions in the final edit. A result that sounds clean alone can change the atmosphere of the scene.
Know the limits
An estimated separation is not the original clean recording
The outputs are estimates from the mixed file. Noise may remain, wanted audio can change, and a word obscured or distorted at recording time cannot be guaranteed recovered. Compare another microphone or clean take first when available.
The service processes uploaded recordings and returns separated WAV audio, not a live filter or an edited video. Use your audio or video editor for final placement and check the finished export. AI processing requires an account and sufficient credits; review the displayed cost for the WAV download.
How do I remove background noise from an audio recording?
Upload a supported recording and describe one audible unwanted source. Review Background audio for the remaining mix, then compare speech, music and noise against the original. If you only need speech, try a speaker prompt and review Isolated sound instead.
Is this the same as a noise-profile filter?
No. This service estimates a described source and its remainder. A noise-profile filter learns a sample of relatively steady noise and may suit a different problem; neither method guarantees an unchanged voice.
Can I remove background noise while keeping music?
You can test a specific unwanted sound as the target and listen to Background audio. Some music may change or some noise may remain, especially when they overlap. Compare instrument attacks and tails with the original.
Does it work on a live microphone or call?
No. It processes an uploaded recording; it does not filter a live microphone feed. Change the recording setup or use a suitable live-processing tool for the next call.
Is the audio cleanup free?
AI processing requires an account and sufficient credits; the results screen shows the WAV download cost. The separate local noise-gate utility is free, but it reduces sound by level and does not identify a noise source.
Start with a sound you know
Bring your own recording.
Describe one source. Listen to both tracks. Keep what works for your project.