Background Noise Remover
Reduce steady background hum, hiss, and fan noise
Your files are processed on your device. The first use downloads a processing engine, which your browser then keeps.
Drop an audio fileTap to choose an audio file
or click to browseor tap to choose
Reduce steady background noise
Large files (over 500 MB) may take longer or run out of browser memory
About Background Noise Remover
Reduces steady background noise — hum, hiss, fan noise, air conditioning — from an audio recording, using spectral subtraction at three strengths, plus a fourth Voice mode that applies a small noise-suppression model loaded from this site. Runs entirely in your browser — the audio never leaves your device.
Upload your audio file, choose how aggressive the cleanup should be, and download the result.
Being upfront about what this can and can't do: this handles steady, constant background noise well — the kind of hum or hiss that's always there at roughly the same level. It's not the same as the AI-based voice isolation in tools like Krisp or Adobe Podcast Enhance, which can separate a voice from much more chaotic or changing background noise (like a crowded room). For a steady noise floor, this genuinely helps; for a complex or shifting noisy environment, results will be more limited.
Watch: Background Noise Remover in Action
- Can this remove background noise from a busy, changing environment like a crowded room?
- Not well — it handles steady, constant background noise well (hum, hiss, fan noise), but isn't the same as AI-based voice isolation in tools like Krisp or Adobe Podcast Enhance, which can separate a voice from more chaotic or shifting background noise. For a steady noise floor, this genuinely helps; for a complex noisy environment, results will be more limited.
- What do Light, Standard and Aggressive actually change?
- They set how many decibels of noise the filter tries to remove, 12, 20 or 35, and how far above your file's measured noise floor sound still counts as noise, 8, 13 or 20 dB. Higher settings clean more but risk dulling the voice.
- When should I use the Voice level?
- Choose it for speech with changing background noise like fans cycling on, traffic or keyboard clicks. It uses a small RNNoise voice model, loaded from this site, and is not suited to music, which can sound damaged by it.
- How does the tool know how loud the noise is?
- It decodes the file in your browser and averages the power of the quietest quarter of short 50 millisecond windows, ignoring digital silence. That value sets the filter's noise floor, limited to the range -80 to -20 dB, so the setting is tuned to each file.
- What does the dB figure under the player mean?
- After processing, the tool compares the same quietest 10 percent of 100 millisecond stretches before and after. The drop in power there is shown as the reduction. It describes the pauses, not the speech, so it is an indication rather than a guarantee.
- Why do I hear a watery or robotic sound afterwards?
- Strong spectral reduction removes some of the voice along with the noise. Go back a level, and avoid Aggressive on thin recordings. If the noise is not steady, try the Voice level or record again in a quieter space.
How to use Background Noise Remover
Drop an audio file, choose a strength button, then press Remove Noise. The choices are Light, Standard, Aggressive and Voice. Light, Standard and Aggressive remove steady hiss, hum and fan noise by different amounts; Voice is meant for spoken recordings where the noise changes, such as keyboards or traffic, and it is not for music.
The tool first measures your file's own noise level, so you do not set a threshold. When it finishes, a line under the player says how many dB the quiet parts dropped, or that little steady noise was found. Compare with the preview, then download denoised_ followed by your file name. Standard is selected at the start.
When to use it
Use it for a podcast, interview or voice memo with a constant hiss, laptop fan, air-conditioning rumble or electrical hum underneath the speech. The Voice level is a better first try for a spoken recording with noise that shifts, such as street sounds through a window or a clicking keyboard.
When not to use it
It cannot separate overlapping voices, remove a dog bark or a passing car cleanly, or rescue a recording where the noise is as loud as the speech. For music, avoid the Voice level. For clipped or distorted audio, noise reduction cannot repair the damage; re-recording closer to the microphone will help more.
Tips for best results
Start with Standard and only go to Aggressive if the noise is still obvious, because strong settings can make voices sound thin or watery. A recording with a second or two of room tone at the start gives the measurement a clean sample. If the reported drop is tiny, the file likely had little steady noise, so another pass will not help.
Limitations
The three spectral levels target steady noise only. For them, the noise floor is taken as the average power of the quietest quarter of 50 ms windows, set a fixed distance above it for each level, and results vary with the recording. If your browser cannot decode a file for that measurement, a default level is used. Only the first two channels are measured, and output follows the source format.
Privacy
Your recording is decoded by the browser's own audio system to measure noise and again to check the result, and the cleaning runs in a local engine, so nothing is uploaded. Two things are fetched: the engine, about 25 MB from a public CDN, and, only for the Voice level, a model file of roughly 300 KB from this site's own server. That request carries no audio.
Next: Silence Remover
Cut dead air and long pauses
More from Audio Tools: