NoiseBG · your audio never leaves this page

Remove Background Noise

Hiss, hum, fans, room tone — learned from your recording and subtracted from it, right here in the browser. A/B the result, read the measured difference, keep the WAV.

Fig. 1 — noisy in · clean out0 bytes uploaded
Drop an audio file here, or tap to choose (MP3 / WAV / OGG / M4A / FLAC)
Before

Recording fresh? Leave a beat of silence before you speak and pick the second option — a clean noise sample is the whole game.

70%
−30 dB

Higher strength eats more noise but can make voices watery. Leaving a whisper of floor sounds more natural than digital black silence.

What this is (and isn't): spectral gating — the technique behind Audacity's noise reduction, tuned and run in your browser. Brilliant on steady noise: hiss, hum, fans, AC, room tone. It is not a dialogue-isolating AI: a dog barking mid-sentence is a sound, not a noise, and anything that promises to erase it in one click is overselling. We'd rather you know that before you press the button.

How to get the best clean

  1. Feed it the noise. The tool learns what to remove from the quiet moments. If you're recording fresh, hold one second of silence before speaking, then use the "first 0.7 s is noise" profile — that sample is worth more than any slider.
  2. Start at 70%. Clean, then A/B. If the voice sounds underwater, back the strength off; if hiss survives, push it up or lower the residual floor. Two passes at moderate strength beat one pass at maximum.
  3. Export the WAV. Lossless out, convert to MP3 afterwards if you need small — compressing before cleaning bakes the noise in.

What's on the bench

Learns your noise

Auto mode finds the quietest slices of your file and fingerprints them; or mark the head of the file as noise-only for surgical profiling.

Measured, not vibes

The readout shows your noise floor before and after in dBFS — a number you can check, not an adjective we chose.

A/B at the tap

Flip between original and cleaned mid-playback. If the cure sounds worse than the disease, you'll know in two seconds.

Strength you control

A slider from gentle to aggressive plus a residual-floor control — the two knobs that matter, without a mixing degree.

Record right here

No file yet? Record from the mic, clean it, export — one page, nothing installed.

Private by architecture

Others promise not to peek at your uploads. We can't peek: there is no server. The difference matters for voice notes that were never meant to travel.

FAQ

Does my audio get uploaded?

No — structurally no. This site is static files; there's no server that could receive audio. Everything runs in your browser and works offline once loaded.

What noise can it remove?

Steady noise: hiss, hum, fans, AC, tape noise, room tone. Not one-off sounds — a cough, a dog, another voice. Those are jobs for an editor's scissors (our trimmer), not a spectral gate.

How much does it remove?

Typically 15–30 dB of floor reduction on steady noise at default strength. The before/after readout shows your actual number.

Why WAV export only?

A good MP3 encoder is a heavyweight dependency we'd have to ship to you. WAV is lossless and everything accepts it — convert after cleaning if you need small files.

File limits?

10 minutes per file — a memory honesty limit, not an upsell. Longer recordings: split, clean, rejoin.

Really free?

Your device does the computing, so there's nothing to bill you for. No account, no watermark, no per-minute meter.

From the bench notes

The rest of the bench