Take fan noise out of a meeting recording
Medium processing on a constant computer-fan hiss brings voices forward.
Guide
Audio Denoise reduces the fairly steady background noise in a recording — fan noise, electrical hum, air-conditioning rumble. The panel offers three afftdn frequency-domain strengths, Light, Medium and Strong, and starts on Medium with “Also filter low-frequency hum below 80Hz” switched on by default.
Updated 2026-09-103 min read
Audio Denoise reduces the fairly steady background noise in a recording — fan noise, electrical hum, air-conditioning rumble. The panel offers three afftdn frequency-domain strengths, Light, Medium and Strong, and starts on Medium with “Also filter low-frequency hum below 80Hz” switched on by default.
This implementation is not AI speech enhancement and does not learn from a noise sample; it runs ffmpeg's FFT denoising filter over the whole recording. The stronger the setting, the more likely it is to shave off detail in the voice as well, and the interface states plainly that the Strong tier can leave audio sounding muffled.
| Input | Output | Notes |
|---|---|---|
Indoor voice-over WAV with constant computer fan noise |
Medium afftdn + 80 Hz high-pass |
Listen first for sibilants and word endings that have been weakened |
Bass recording with low-frequency content |
Light afftdn with the 80 Hz high-pass off |
Avoid filtering out the low end you want to keep |
Medium processing on a constant computer-fan hiss brings voices forward.
Switch on the 80 Hz high-pass to push the low hum down, then pair it with light or medium afftdn.
Denoise lightly first, then remove silence and normalise loudness.
No, not reliably. Another speaker is not steady noise and their spectrum overlaps the target voice, so proper source separation or manual editing is needed.
More aggressive frequency-domain suppression cuts voice detail along with the noise — the typical trade-off of this filter.
The recording is processed locally in the browser by ffmpeg.wasm running afftdn / highpass and is never uploaded; the first use of the workbench downloads about 31–32 MB of ffmpeg core from the site's asset domain or jsDelivr.
Updated 2026-09-10
Three strengths of afftdn spectral denoising, plus an optional 80 Hz high-pass to cut low hum
Everything runs locally in your browser; files are never uploaded to a server