Audio Noise Reducer

Why your recording sounds noisy — and how to fix it

You hit record in a real room, not a treated studio, so the mic also captured everything you didn't want: the steady hiss of the preamp, a low 60 Hz hum from the mains, the drone of an air conditioner, and the general "air" of the space. That background bed is broadband noise — energy spread across many frequencies at a low, constant level — and it sits underneath your voice for the entire take. This tool removes it. Upload a noisy MP3, WAV, M4A, AAC, FLAC, OGG or WebM file, pick a strength, and download a cleaned version, with a measured before-and-after noise floor in dBFS so you can prove the hiss actually dropped instead of just hoping it did.

What the noise floor is

The noise floor is the level of the quietest part of your recording — the room tone you hear between words — measured in dBFS, where a lower (more negative) number means a cleaner, quieter background. A clean voice recording usually sits around −60 dBFS or lower; anything above roughly −50 dBFS has audible hiss. Lowering that floor is exactly what noise reduction does, which is why this tool measures it before and after and shows you the difference in plain decibels.

How the FFT denoiser works (and why it won't gut your voice)

Under the hood this tool uses ffmpeg's afftdn filter — an FFT denoiser. It chops the audio into short, overlapping frames, runs a Fast Fourier Transform to break each frame into its individual frequencies, estimates how much of each frequency is noise versus signal, and then pulls down only the noise component before rebuilding the waveform. Because it works frequency-by-frequency, it can lift a constant hiss out from under speech while leaving the speech itself largely intact — something a simple volume gate or low-pass filter cannot do.

The trade-off is strength. Remove a little and the result sounds natural; remove too much and you start eating into the voice itself, producing the watery, underwater, slightly robotic artifacts everyone recognizes from over-processed audio. That is why Medium (12 dB) is the default: it clears typical voice-recording hiss without obvious artifacts. Light (6 dB) is for subtle hiss you want gone without touching the tone; Strong (24 dB) is for heavy hum or loud backgrounds, and you should listen back for artifacts afterwards. The optional Remove low-frequency rumble toggle adds an 80 Hz high-pass first, clearing mic handling, footsteps and AC rumble that denoising alone leaves behind.

The measure → clean → normalize workflow

Cleanup is one step in a chain. First, measure the noise floor to see how bad it is and whether cleanup is even needed. Then clean it here. Finally, normalize the loudness so the now-quieter file hits the −16 LUFS (podcast) or −14 LUFS (YouTube) target your platform expects. Do it in that order — normalizing before you denoise just makes the noise louder too, and then you are cleaning a hotter signal.

Noise reducer vs Background Voice Remover

These solve different problems. This tool removes broadband noise — hiss, hum, fans, air. If instead you are fighting background voices or chatter — other people talking behind your subject — that is a source-separation problem, not a noise-floor problem, and the Background Voice Remover is the right tool. Running the denoiser on background speech will not remove the talking; it will just make the whole recording sound processed.

Limits & privacy

Without an account you can clean one file up to 10 MB and 5 minutes long; a free account raises that to 100 MB and 30 minutes. The output is a lossless WAV, because re-compressing a freshly cleaned signal back to MP3 would stack a second round of artifacts on top of the ones you just removed — if you need a smaller file afterwards, convert it as a separate step. Your file is processed securely on our servers and automatically deleted after you download it, and no account is required to use the tool.