Air conditioning, a fan, a computer in the corner — the steady noise under your voice is measured first, then removed. You get the real reduction in decibels, not a made-up percentage, and the speech level stays where it was.
−8.6 dB
−11.8 dB
−22.1 dB
−43.5 dB
Most free denoisers apply the same fixed setting to every file. That is why they do almost nothing to a quiet recording and destroy a noisy one. This tool reads your file before it touches it.
The file is scanned second by second and the quietest passages are taken as the real noise floor. Every setting is then aimed at that measurement.
Soft through Brutal do not just change a label. On our test file they removed 8.6, 11.8, 22.1 and 43.5 decibels of noise respectively.
Across all four settings the speech level moved by less than one decibel. If a denoiser is quietening your voice, it is doing the wrong job.
An optional 80 Hz high-pass takes out electrical hum, desk thumps and handling rumble before anything else is measured.
Render ten seconds with your chosen settings and play it against the original, side by side, before processing the whole file.
The finished file is measured again, so the decibel figure on the result screen is a real comparison between two files rather than an estimate.
Drop in the recording. It is checked with FFprobe, so a renamed file that is not really audio is refused straight away.
You see the measured noise floor, the speech level and the gap between them before choosing anything.
Ten seconds with your settings, played next to the original. Change the strength and render again until it sounds right.
Process the whole file and download it as MP3, WAV or M4A. Your upload is deleted the moment it finishes.
Constant, broadband and completely predictable — the ideal case.
A steady tone sitting under the whole recording comes out cleanly.
Preamp hiss and 50/60 Hz hum, especially with the high-pass switched on.
One-off sounds are not steady, so there is no profile to subtract.
Separating two voices is a different problem, and no filter of this kind does it.
Limits. 200 MB or 3 hours per file, ten files per visitor per day, and two jobs at once with a short queue behind them. These limits are what let the tool stay genuinely free.
You only noticed the hum in the edit. It is steady, so it comes out.
Fan noise on one side of a conversation is the classic case for this tool.
Broadband hiss under an archive recording responds well to the stronger settings.
The high-pass removes the low-frequency rumble that eats your headroom.
Stronger is not always better. Brutal removed 43.5 dB on our test file, which is enormous. On a recording where the noise sits close to the voice, that much subtraction starts to thin out breaths and the quiet ends of words. Start at Normal, preview, and only go further if you can hear that you need to.
Do not normalise here. There is an option to bring the output to −16 LUFS, and we measured what it costs: signal-to-noise improved by 11.1 dB with it switched off and only 2.2 dB with it on. Normalising after cleaning applies gain to whatever noise is left, and its dynamic stage lifts the quiet passages most. Clean here, then normalise in the Podcast Audio Studio, which measures twice and corrects itself.
Not at Soft or Normal. The stronger settings subtract more of the spectrum and can take a little life out of breaths and sibilance, which is why there is a preview.
On our test file, between 8.6 dB at the gentlest setting and 43.5 dB at the strongest. Your result depends on how steady your noise is; a constant hum comes out far better than a busy street.
No. Speech level moved by less than one decibel at every setting in testing. If you also want a consistent publishing level, use the Podcast Audio Studio afterwards.
No, and that is deliberate. It is spectral subtraction with FFmpeg, aimed at a noise floor measured from your own file. There is no model, no external service and no cost.
It silences passages below a threshold, which removes more between sentences but can clip the ends of words. It is off by default for exactly that reason.
The upload is deleted as soon as processing finishes and the result within two hours. No account, no database, nothing kept.
Upload, read the noise profile, preview ten seconds and download.
Open the Noise Remover →This page belongs to Bubixo Voice and Speech Tools, where every tool in the category is listed. These are the ones people usually open next.