What this actually removes
This tool targets the steady, constant noise that sits under most home recordings: air-conditioner hum, computer fan whir, mains buzz, tape hiss, and the broad "room tone" you only notice once it is gone. It runs FFmpeg's adaptive frequency-domain denoiser (afftdn), then trims sub-80 Hz rumble and ultrasonic fizz with a high-pass and low-pass filter, and finishes with a loudness-normalize pass so the result lands at a consistent broadcast level. It will not magically erase a dog barking, a door slamming, or someone talking over you — those are transient, non-stationary sounds that need a spectral editor, not a single denoise pass.
How the three strength settings differ
The Light / Medium / Strong buttons change the noise-floor estimate the denoiser works from. Light leaves more of the original ambience intact and is the safest choice when the noise is mild and you care about keeping the voice natural. Strong pushes the floor up so more of the constant hiss and hum is gated out — great for a noisy fan or a hummy USB mic, but lean on it and quiet speech can start to sound thin or "underwater." Medium is the sensible default for most laptop-mic and phone recordings. When in doubt, run Medium first, listen, and only step up to Strong if there is audible noise left.
Got a video, not an audio file?
This tool is audio-only by design — it outputs an MP3, not a re-muxed video. If your noise is inside a screen recording or a talking-head clip, pull the soundtrack out first with our video to audio tool, clean the resulting file here, and then mux the cleaned audio back over your video in an editor. Keeping the steps separate means each pass does one job well and you can audition the denoise on its own before committing it to the final cut.
Why FFmpeg instead of an AI denoiser
The big AI denoisers (the kind baked into paid meeting apps and cloud studios) are genuinely better at separating a voice from chaotic background sound — but they cost money, require an account, and ship your raw audio to someone else's server. afftdn is an old-school, deterministic, frequency-domain filter: it is free, it is private, and it runs on your own machine with nothing uploaded. For the common case — a single voice over steady hiss or hum — it gets you 90% of the way for 0% of the cost or signup friction. If you would rather not fight noise after the fact at all, Clipy captures cleaner mic audio at the source and auto-transcribes every recording, so there is less to clean up to begin with.