Remove Background Noise
- No Sign-up
- 100% Free
- Zero Server Data
Wind on the microphone, the hiss of a cheap preamp and the hum of mains electricity are three different problems, and no single filter removes all three. This handles each one the way it needs handling: a slope that cuts the rumble below the voice, narrow notches on the mains frequency it measures rather than guesses, and a noise spectrum learned from your own recording and subtracted from it. When it has finished it tells you, in decibels, how much quieter the background became. Everything happens in your browser.
1 · Your recording
2 · What to remove
Start with Balanced. Strong removes more and can leave a voice sounding thin, which is what the setting is trading away.
Frequently asked questions
Is my recording uploaded anywhere?
No. Your browser decodes the file, does the arithmetic and writes a new one, all in the tab. The site’s security policy sets connect-src ‘none’, so the browser blocks network requests from these pages, so the audio could not be sent even if the code tried. That matters more here than on most pages, because the recordings people want cleaned up are interviews, lectures, voice notes and calls.
How does it know what the noise sounds like?
It measures it. Noise is present the whole time, so the quietest level each frequency ever reaches during the recording is what that frequency's noise sounds like, and there is no need for anyone to find a silent passage and select it. That estimate is then subtracted from every moment of the recording.
What is the difference between the three?
Where the noise sits. Wind and rumble are almost entirely below the lowest note a voice makes, so a filter removes them and takes nothing with it. Mains hum is a handful of exact frequencies, so narrow notches remove it and leave everything between them alone. Hiss is spread across the whole range and sits on top of the voice rather than beside it, which is why it needs the measured spectrum and why it is the one that costs something.
Why does it ask about 50 or 60 Hz nowhere?
Because it works it out. Mains is 50 Hz in Thailand, Europe and most of Asia and 60 Hz in the Americas and Japan, and a notch at the wrong one removes a piece of the voice and leaves the hum exactly as it was. The page measures both and reports which it found.
Will it make a voice sound strange?
Balanced usually does not. Strong can: pushing a subtraction harder removes more noise and starts taking the quiet edges of speech with it, which is heard as a thin or underwater quality. The setting is that trade, and it is why Balanced is the default rather than the strongest option.
Does it work on music?
Less well, and honestly so. The method finds the noise floor by looking for the quietest moment at each frequency, and a note held longer than a second and a half looks exactly like noise by that measure. Speech moves constantly, which is why it survives. This is a tool for recordings of people talking.
Why M4A and not MP3?
Because the browser will not encode MP3. AudioEncoder was asked directly and reported AAC and Opus supported and MP3 not, so offering it would mean shipping an encoder for a format whose advantage over M4A is familiarity rather than quality or compatibility. M4A plays in everything. WAV is there for software that will not open one.