Noise Remover
Learn the noise floor of a recording, then gate it out band by band
100% In-Browser · Private & Instant · Free Forever
Your file is decoded, processed, and re-encoded in this browser tab. Nothing is uploaded to an OfflineTTS processing API.
Interactive Workspace
Noise Remover
Remove steady background noise from a recording: fan hum, air conditioning, computer hiss, traffic rumble, and general room tone. The tool learns the noise floor from the quietest half-second of your file, then attenuates every frequency band that sits below it.
How spectral gating works: the audio is analysed in overlapping windows, and each window is broken into frequency bands. Bands whose energy is below the learned noise floor are turned down, with a smooth curve rather than a hard on/off switch. Because the decision is made per band and per window, a voice can be loud in one band while a fan hum is simultaneously suppressed in another.
Voice focus: with this on, the 250 Hz–4 kHz range — where speech intelligibility lives — is treated more gently than the rest of the spectrum. Noise below and above the voice is reduced harder, which is what you want for speech and not for music.
The rumble filter: a steep high-pass at the frequency you choose. Mains hum, desk vibration, and traffic all live below 100 Hz, and removing that band before gating reduces the work the gate has to do.
Set expectations: this is a spectral gate, not a trained speech-enhancement model. It works extremely well on steady noise and much less well on intermittent sounds such as keyboard clicks, door slams, or a second speaker. Pushing the strength too high produces a watery, metallic artefact often called "musical noise" — when that appears, back the strength off. Decoding, processing, and encoding all run in this browser tab. The file you select is not uploaded, and the rendered audio is not stored on any server.
How It Works
Load a noisy recording
Add the file that needs cleaning up.
Read the noise floor
The analysis panel reports the measured floor level and where it was found.
Set strength and mode
Raise the reduction until the noise drops, keeping voice focus on for speech.
Render and export
Compare against the original, then download WAV or MP3.
Features
Learned Noise Floor
Automatically finds the quietest window and profiles its spectrum.
Voice Focus
Protects the 250 Hz–4 kHz speech band from over-gating.
Rumble Filter
A steep high-pass that removes hum and traffic before gating.
Measured Before After
Reports the detected noise floor so you can judge the settings.
Why the Quietest Window Matters
Spectral gating needs to know what the noise looks like before it can remove it. This tool measures the quietest half-second of the file and treats that spectrum as the noise profile. The assumption is that the quietest moment contains mostly noise and very little signal, which is true for interviews, voice notes, and narration with natural pauses.
That assumption breaks down in two situations. Continuous music has no quiet window, so the learned profile includes musical content and the gate starts removing parts of the track. A recording with no pauses at all — a single unbroken take with constant background noise — gives the profile nothing clean to measure. In both cases, the result is a gate doing something reasonable with an unreliable reference.
If you have a recording with a usable silent section, the result will be noticeably better than one without. Ten seconds of room tone recorded before the take is worth more than any amount of adjustment afterwards.
Strength, Artefacts, and Realistic Expectations
The strength control sets how far above the learned noise floor a band must sit before it is left alone. Low values are conservative and preserve detail; high values remove more noise but start to carve into quiet parts of the voice, producing a hollow, processed sound. The useful range for most material is between 40% and 65%.
Noise removal is inherently a trade-off. Every decibel of noise removed is a decibel of signal that was sitting at a similar level, and no algorithm can distinguish the two perfectly. Tools that claim otherwise are either using a trained model with the artefacts that come with it, or overstating the result. This one is honest about being spectral gating.
Every effect here is deterministic: the same file and the same control values produce the same output. That means you can reproduce a result later, but it also means the tool will not guess what you intended — check a short excerpt before rendering a long file, and keep the original.
Keep the original recording. If a cleaned version turns out to have removed too much, the only way back is to re-render from the source with gentler settings — so do not overwrite the file you started from.
Related Tools
Volume Booster
Make a quiet recording louder and normalise inconsistent levels
Audio Equalizer
Ten bands from 32 Hz to 16 kHz, with presets and a spectrum view
Voice Changer
Sixteen character presets, each one still fully editable
Audio Converter
Change sample rate and channel count, normalize, and re-export
Noise Remover — FAQ
What kinds of noise can this remove?
Steady, broadband noise: fan and air-conditioning hum, computer hiss, tape noise, traffic rumble, and constant room tone. It works best when the noise does not change much over time.
What can it not remove?
Intermittent sounds that overlap the voice: keyboard clicks, door slams, coughs, and a second speaker talking. Those share the same frequency bands and the same time windows as the speech, so a spectral gate cannot separate them. Removing them requires editing, not filtering.
What is "musical noise" and how do I avoid it?
When too many frequency bands are gated at once, the remaining signal develops a watery, warbling quality. It comes from isolated bins surviving the gate while their neighbours are suppressed. The fix is to reduce the strength and to use voice focus, which gates the speech band more gently.
Should I clean the audio before or after boosting the volume?
Clean first, then boost. Boosting raises the noise floor along with everything else, so a file that was cleaned at its original level stays cleaner after gain than one where the gain came first.
Does it work on music?
It works, but the result is rarely what you want. Music occupies the whole spectrum continuously, so there is no quiet window to learn a noise floor from, and gating removes quiet musical detail along with the noise. This tool is designed for speech.
More Local Audio Tools
Every audio utility here runs in the browser tab. Browse the full set of editing, conversion, and measurement tools.