Denoise Audio

Remove Background Noise from Audio — Fastest Free

QUICK ANSWER

The fastest free audio noise remover online. Drop a recording and a server-side ffmpeg afftdn + high-pass + low-pass + loudnorm chain strips hum, hiss, fan whir, and room tone, then evens the level to broadcast loudness. Works across MP3, WAV, M4A, AAC, OGG, and FLAC. No signup, no watermark.

  • Native ffmpeg afftdn pipeline
  • Light / Medium / Strong
  • Loudness-normalize to -16 LUFS
  • MP3 / WAV / M4A / OGG / FLAC in
  • No watermark, no signup
  • Up to 500 MB

Files are deleted from our server within 24 hours.

Clipy

Editing clips just to share them?

4,000+ people record and share with Clipy — free forever, no watermark.

  • Record, trim, and share in one tab
  • Instant link the moment you stop
  • Agent-readable links: AI summary, key moments, transcript
  • Free forever — no watermark, no install
Yes, record & share free

No credit card. Viewers never need an account.

Why this tool exists

  • Files upload to the nearest Backblaze B2 POP via presigned URL.
  • Native ffmpeg afftdn + loudnorm on the server. No watermark, no signup.
  • Source and output are deleted within 24 hours.
40+ conversions and counting
Trusted by creators at startups, agencies, and Fortune 500 teams.
Free forever — no signup, no watermark.

Why this is fast

Native server ffmpeg with real SIMD and threading runs the denoise + normalize chain roughly 4–8x faster than browser WebAssembly on the same audio. afftdn is a real-time-capable frequency-domain filter, so even a 60-minute podcast finishes in well under a minute of CPU time. Uploads land at the nearest Backblaze B2 point of presence via a presigned URL, the cleaned MP3 comes back through Bunny CDN from the same edge.

What this tool does, exactly

The filter chain is afftdn=nf=<floor>, highpass=f=80, lowpass=f=12000, loudnorm=I=-16:TP=-1.5:LRA=11 followed by LAME at VBR quality 2 (~190 kbps). afftdn is ffmpeg's adaptive frequency-domain denoiser — it estimates the noise floor and gates everything below it. The highpass at 80 Hz removes sub-bass rumble (HVAC, mic-stand thumps). The lowpass at 12 kHz drops ultrasonic fizz and high-frequency hiss. loudnorm targets the EBU R128 broadcast standard (-16 LUFS integrated, -1.5 dBTP, 11 LU range), so the output sits at a comfortable, consistent level.

How afftdn beats arnndn for one-shot cleanup

Two reasons. First, afftdn is bundled with ffmpeg and needs no model file — arnndn requires a trained RNNoise model (.rnnn) that has to ship with the binary, and the open models are tuned for specific noise profiles. Second, afftdn is adaptive: it estimates the noise floor from your file and gates frequencies below it, which works on whatever mix of hiss, hum, and fan whir your recording happens to contain. arnndn is better at heavy chaotic backgrounds when its model matches your noise, but when it does not the cleanup can sound spectrally strange. For the common case — a single voice over steady noise — afftdn at the Medium (-25 dB) noise floor is the right default.

Sister tools

Got noise inside a video, not a standalone audio file? Strip the soundtrack first with video to audio converter, clean it here, then mux the cleaned audio back over the video in an editor. Loudness is the only problem? Normalize video audio targets -16 LUFS without touching noise. Trim the cleaned MP3? MP3 cutter. Shrink it for email? Compress audio. Check the mic before the next recording to avoid the cleanup altogether? Microphone test online. Recording a fresh take in the browser? Voice recorder captures straight to MP3.

Common questions

Will this remove someone talking in the background or a dog barking?

No. afftdn removes steady, constant noise &mdash; hum, hiss, fan whir, room tone. Transient, irregular sounds like speech, barks, or door slams are non-stationary and need a spectral editor (Audacity&apos;s noise reduction with a manual selection, or an AI separation tool like ElevenLabs Voice Isolator), not a single denoise pass.

How does afftdn beat arnndn for one-shot cleanup?

Two reasons. First, afftdn is bundled with ffmpeg and needs no model file &mdash; arnndn requires a trained RNNoise model (.rnnn) that has to ship with the binary, and the open models are tuned for specific noise profiles. Second, afftdn is adaptive: it estimates the noise floor from your file and gates frequencies below it, which works on whatever mix of hiss, hum, and fan whir your recording happens to contain. arnndn is better at heavy chaotic backgrounds when its model matches your noise &mdash; but when it does not, the cleanup can sound spectrally weird. For the common case (a single voice over steady noise), afftdn at -25 dB noise floor is the right default.

Why does the output come back as an MP3 even if I uploaded a WAV?

We re-encode to MP3 at LAME VBR ~190 kbps so the result is small and universally playable. If you specifically need lossless WAV or FLAC out, this is not the tool for that yet &mdash; the V1 spec pins one output container per slug. A WAV output toggle is on the roadmap.

My cleaned audio sounds thin or 'underwater' &mdash; what happened?

That is over-denoising. Drop from Strong to Medium or Light. The stronger the setting, the more of the signal near the noise floor gets gated out, which can hollow out quiet speech. Pick the lightest setting that still kills the noise you can hear.

Does it make my recording louder too?

Yes &mdash; after denoising it runs a loudness-normalize pass to an integrated -16 LUFS target with a -1.5 dB true-peak ceiling, which is the standard target for spoken-word and podcast audio. So the output is both quieter in noise and more consistent in volume.

How do the three strength settings differ?

Light leaves more of the original ambience intact (afftdn noise floor -30 dB) and is the safest choice when noise is mild and you care about keeping the voice natural. Medium (-25 dB, the default) handles most laptop-mic and phone recordings. Strong (-20 dB) pushes the floor up so more of the constant hiss and hum is gated out &mdash; great for a noisy fan or hummy USB mic, but lean on it and quiet speech can start to sound thin.

What happens to my file after the cleanup?

The source uploads to the nearest Backblaze B2 point of presence via a presigned URL, the afftdn + loudnorm chain runs on our server, the cleaned MP3 is delivered through Bunny CDN, and both files are deleted within 24 hours. No signup, no watermark.

Tool not working the way you expect?

Tell us what broke — we resolve every feedback within 24 hours.

Send feedback →

More free tools

Doing this for an AI workflow?

Clipy recordings are agent-readable out of the box — one link carries the video, keyframes, and transcript.

Clipy for AI agents