Loading tool…
Computer fans, air conditioners, mic hiss and traffic rumble make good recordings tiring to listen to. Mark a quiet stretch, let the tool learn how loud the noise is, compare a cleaned preview with the original, then export the whole file without uploading anything.
Processing locally on your device
Overview
The main stage is FFmpeg’s afftdn filter, a frequency-domain denoiser. It splits the audio into short overlapping frames, converts each to frequencies with a fast Fourier transform, and turns down the parts of the spectrum that sit near the noise level while leaving louder content, your voice, alone. It runs as FFmpeg compiled to WebAssembly inside your browser. The engine is about 31 MB the first time and is cached after that, so later runs start right away.
Two settings matter most. Reduction, 1–40 dB, sets how far noise is pushed down. The noise floor, −80 to −20 dB, tells the filter how loud the noise actually is, and it makes the biggest difference. When you open a file the tool estimates it automatically from the quietest moments; for better results, drag across a section that contains only background noise and click Learn from selection, which measures that section’s level. In testing on a voice-plus-hiss sample, a floor matched to the noise cut high-frequency hiss by about 14 dB while the voice level stayed the same.
A rumble filter (high-pass at 60, 80 or 120 Hz) removes traffic, handling and air-conditioning rumble that the ear barely notices but that eats headroom. Adapt to changing noise lets the filter follow noise that drifts over time; it is gentler, so leave it off for steady hiss or hum. Extra smoothing adds FFmpeg’s anlmdn non-local-means denoiser after the main pass; it can soften leftover grain but is slow, roughly as long as the audio itself in a browser.
Noise reduction always trades noise for artifacts. Pushing too far creates a watery, warbling sound or dulls the voice, so use the before/after preview: it cleans a 10-second section from the playhead and lets you flip between the original and the result in sync. Steady noise such as hiss, hum and fans is much easier to remove than music, overlapping voices or echo.
| Noise | Character | Best setting here |
|---|---|---|
| Mic or tape hiss | Broadband, high frequencies | Learn floor from a quiet part, 12–20 dB |
| Fan / air conditioner | Steady, broadband | Learn floor, 12–20 dB, rumble filter 80 Hz |
| Low rumble, handling | Below about 80 Hz | Rumble filter 80–120 Hz |
| Slowly changing noise | Drifts over time | Adapt to changing noise on |
| Voices, music, echo | Not steady | Hard to remove; record again if possible |
Open a recording with background noise (audio or video).
Drag across a part with only noise and click Learn from selection, or keep the automatic estimate.
Set the reduction amount and choose a rumble filter.
Move the playhead to a typical spot, click Render preview and compare A and B.
Export the cleaned file as MP3, M4A, WAV, FLAC, OGG or Opus.
Use cases
Remove the steady hum of a computer fan or air conditioner from narration recorded in a spare room, then add the rumble filter to clear desk thumps and traffic.
Clean up phone or camera recordings made in a classroom or meeting room before transcribing them or sharing them with listeners.
Reduce the constant hiss on digitized cassette recordings of family voices or old radio shows. Learn the floor from the leader before the recording starts.
Load a video, clean its soundtrack, and drop the new audio back into your editor to replace the noisy original track.
Pro tips
Answers to common questions about the noise reducer.
Not reliably. Spectral noise reduction works best on steady sounds such as hiss, hum, fans and air conditioning. Separating voices or music needs source-separation models, which is a different kind of tool.
That is “musical noise,” an artifact of aggressive spectral reduction. Lower the reduction amount or set the noise floor a little lower. The goal is unobtrusive noise, not total silence.
It measures the loudness of the part you selected and uses it as the filter’s noise floor. Select a stretch with only background noise, no speech or breaths, for the most accurate setting.
The first time, your browser downloads the FFmpeg engine, about 31 MB. It is cached afterwards, so later previews and exports start immediately, even offline.
The main noise reduction is usually much faster than real time on a laptop. Extra smoothing is far slower and can take about as long as the audio itself, so use it on short clips or leave it off.
No. ToolEnhancer processes files on your own device inside the browser. They are never sent to, stored on or seen by our servers.
Yes. The noise reducer is free to use with no sign-up, no watermarks and no daily limits.