Audio Noise Remover

Remove background noise from audio recordings using spectral subtraction.

Tested tool guide Tested browser tools Checked August 15, 2026

What Audio Noise Remover does and how it behaves

This tool reduces steady background noise (tape hiss, fan hum, room tone) in an audio file using spectral subtraction: it learns the frequency profile of the noise from a quiet stretch of the recording, then subtracts that profile from the rest of the signal. It runs entirely in the browser on a file you choose. The most common surprise is that pushing the reduction strength too far introduces a warbling, robotic artifact known as musical noise, and the technique barely touches non-steady sounds like clicks, traffic, or a second voice.

How the result is produced

Noise profile capture

The tool analyzes a segment identified as noise-only (either auto-detected silence or a range you mark) and computes the average magnitude spectrum across frequency bins using an FFT. This becomes the noise floor estimate: how much energy is expected in each frequency bin purely from background noise, independent of the wanted signal.

Frame-by-frame subtraction

The full recording is split into overlapping short frames, each converted to the frequency domain. The stored noise magnitude is subtracted from each frame's magnitude spectrum, with a floor so bins don't go negative, while the original phase is kept. Frames are inverse-transformed and overlap-added back into a waveform.

Good uses

Cleaning tape hiss or vinyl surface noise out of a digitized cassette or record transfer
Removing constant fan, AC, or electrical hum from a voice memo or podcast segment before editing
Suppressing steady room noise in a field-recorded interview so a transcription pass is easier to follow

Limits and checks

Effective mainly on stationary noise with a stable spectrum (hiss, hum); it does not remove transient sounds like clicks, pops, or a passing car, and won't separate one voice from another talking over it
Higher reduction strength trades noise removal for musical-noise artifacts - a warbling or metallic residue that's often more distracting than the original noise, especially in quiet passages
Output quality depends on picking a noise-only sample segment; if that segment contains any speech or wanted signal, the profile is wrong and the subtraction will dull or distort legitimate audio

Common questions

Can this remove a second person talking in the background?

No. Spectral subtraction targets a steady, learnable noise profile like hiss or hum. A second voice shares too much frequency and time structure with the voice you want to keep, so the tool can't cleanly separate the two - it will smear both instead of isolating one.

Why does the cleaned audio sound a bit warbly or robotic?

That's musical noise, a known side effect of spectral subtraction when the reduction strength is set too high or the noise profile wasn't perfectly stable. Lowering the strength or re-sampling a cleaner noise-only segment usually reduces it, though some residual artifact is normal.

References and verification

The behavioral notes were checked against the browser implementation. Standards and primary references below define the relevant format, formula, or platform behavior.