Remove Background Noise
Clean a recording in the tab with three stated stages: a high-pass filter, a mains notch and a broadband noise gate, each with named parameters.
What this page makes
Applies a one-pole high-pass, an optional 50 or 60 Hz band-stop and a block envelope gate: a filter chain, not a model.
How to use it
Choose a file, set the high-pass cutoff, choose a mains notch or none, and set the gate threshold, attack and release. Press Process; the readout prints the exact chain and the RMS before and after. Play processed or Download processed WAV.
Worked example: a 5.000 s recording at 48 kHz, high-pass 80 Hz, notch 60 Hz, gate −40 dBFS becomes mains hum and rumble attenuated, the gate closes between phrases, and a fan at −35 dBFS stays because it is above the threshold.
Method
Stage one is a one-pole high-pass at the cutoff you set, 80 Hz by default. Stage two is an RBJ band-stop at 50 or 60 Hz with Q = 30. Stage three is a broadband gate over 256-sample blocks: a block whose RMS is below the threshold in dBFS is attenuated, and the gain ramps with the attack and release you set, 10 ms and 120 ms by default. Each channel is filtered independently.
What it refuses
A cutoff outside 20–400 Hz, a threshold outside −80 to −10 dBFS, or an attack or release outside its range is refused with the value named. The page refuses to remove reverberation or to reconstruct speech: it is a gate, not a denoiser model.
Where it runs
The file is decoded and filtered in this tab: nothing is uploaded, no noise profile is learned and no processed copy is stored. The browser requirement is stated on the page: Decoding needs decodeAudioData and playback needs an AudioContext. The three filter stages are plain sample arithmetic in the tab; no model and no network are involved.
The rest of the family
The other audio and recurrence pages are listed under Calculators: tone and binaural generation, the virtual piano, pitch shifting, looping and reversal, the stated noise filter chain, mid/side cancellation, microphone recording and the closed-form recurrence solver.
Applies a one-pole high-pass, an optional 50 or 60 Hz band-stop and a block envelope gate: a filter chain, not a model.
Stage one is a one-pole high-pass at the cutoff you set, 80 Hz by default. Stage two is an RBJ band-stop at 50 or 60 Hz with Q = 30. Stage three is a broadband gate over 256-sample blocks: a block whose RMS is below the threshold in dBFS is attenuated, and the gain ramps with the attack and release you set, 10 ms and 120 ms by default. Each channel is filtered independently. The file is decoded and filtered in this tab: nothing is uploaded, no noise profile is learned and no processed copy is stored.
Worked example: a 5.000 s recording at 48 kHz, high-pass 80 Hz, notch 60 Hz, gate −40 dBFS becomes mains hum and rumble attenuated, the gate closes between phrases, and a fan at −35 dBFS stays because it is above the threshold.
The readout prints the exact chain and the RMS before and after, so the result can be checked rather than trusted.
How to use the remove background noise
Choose a file, set the high-pass cutoff, choose a mains notch or none, and set the gate threshold, attack and release. Press Process; the readout prints the exact chain and the RMS before and after. Play processed or Download processed WAV.
The file is decoded and filtered in this tab: nothing is uploaded, no noise profile is learned and no processed copy is stored.
- Domain: One decodable audio file; high-pass 20–400 Hz, notch 50 or 60 Hz with Q 30, gate −80 to −10 dBFS, attack 0–200 ms, release 0–1000 ms.
What the page proves
Applies a one-pole high-pass, an optional 50 or 60 Hz band-stop and a block envelope gate: a filter chain, not a model.
Stage one is a one-pole high-pass at the cutoff you set, 80 Hz by default. Stage two is an RBJ band-stop at 50 or 60 Hz with Q = 30. Stage three is a broadband gate over 256-sample blocks: a block whose RMS is below the threshold in dBFS is attenuated, and the gain ramps with the attack and release you set, 10 ms and 120 ms by default. Each channel is filtered independently.
- Browser requirement: Decoding needs decodeAudioData and playback needs an AudioContext. The three filter stages are plain sample arithmetic in the tab; no model and no network are involved.
- The page states its domain: One decodable audio file; high-pass 20–400 Hz, notch 50 or 60 Hz with Q 30, gate −80 to −10 dBFS, attack 0–200 ms, release 0–1000 ms.
What it refuses, and what it leaves out
A cutoff outside 20–400 Hz, a threshold outside −80 to −10 dBFS, or an attack or release outside its range is refused with the value named. The page refuses to remove reverberation or to reconstruct speech: it is a gate, not a denoiser model.
- Not the right page for: removing a fan or traffic noise above the threshold
- Not the right page for: removing reverberation
- Not the right page for: restoring speech that the gate has already cut
- Reported as a named state: a threshold above the noise leaves the noise in place
- Reported as a named state: a high-pass too high thins the voice
- Reported as a named state: a short release cuts word tails
What to Expect
Best for
- removing mains hum from a voice memo
- cutting room rumble below a spoken phrase
- silencing gaps between sentences
Not ideal for
- removing a fan or traffic noise above the threshold
- removing reverberation
- restoring speech that the gate has already cut
What this tool keeps
- every sample outside the filtered bands; the parameters are applied, not adapted
What may need cleanup
- set the threshold just above the noise floor, not above the quietest word
- prefer a longer release if word tails are being cut
Common errors
- a threshold above the noise leaves the noise in place
- a high-pass too high thins the voice
- a short release cuts word tails
Example use cases
- clean a voice memo recorded near a mains cable
- remove low rumble before transcribing
- silence the gaps in a narration
Sample input
a 5.000 s recording at 48 kHz, high-pass 80 Hz, notch 60 Hz, gate −40 dBFS
Sample output
mains hum and rumble attenuated, the gate closes between phrases, and a fan at −35 dBFS stays because it is above the threshold
Who this is for
- podcasters cleaning a rough take
- anyone with a noisy voice memo
Frequently asked questions
Is my data sent anywhere?
The file is decoded and filtered in this tab: nothing is uploaded, no noise profile is learned and no processed copy is stored.
Which browser API does this page need?
Decoding needs decodeAudioData and playback needs an AudioContext. The three filter stages are plain sample arithmetic in the tab; no model and no network are involved.
What can I download?
Download processed WAV writes the filtered channels as 16-bit PCM.
Is this an AI denoiser?
No. It is three named stages: a one-pole high-pass, an RBJ band-stop and an envelope gate. No model is loaded, no noise profile is learned and nothing is uploaded.
Why does my voice sound thin after processing?
The high-pass removes everything below its cutoff. At 80 Hz that is rumble, but pushing it to 300 Hz removes the lower part of a voice; the page names the cutoff so you can lower it.
Why are word endings cut off?
The gate closed during a quiet ending and the release is too short. Raise the release or lower the threshold; the page prints both values it used.
Build custom tools for your store
Join the SDK program to create useful ecommerce utilities for your store.