Noise Filtering: Remove Background Noise From Audio
Clean background noise out of a recording in your browser: hiss, hum and room rumble go, the voice stays.
This tool estimates the speech in each moment of a recording and rebuilds the signal without the noise around it, rather than gating or ducking the quiet parts. It works frame by frame over a short-time spectrum, so steady offenders, hiss, mains hum, fans, room rumble, drop away while the voice keeps its timing and tone. Three model sizes are offered: the smallest is near-instant, the largest suppresses hardest on difficult material. Processing stays on your device and the result exports as 48 kHz WAV.
How to Use It
- Upload your audio file and wait for the local preview to load.
- Adjust the settings: Adjust the available settings and run the tool locally in your browser.
- Download the result: Download the processed file directly to your device.
Frequently Asked Questions
What kind of noise does this remove?
Steady, broadband interference is the target: tape or preamp hiss, mains hum, air-conditioning, fans, room rumble and similar constant backgrounds. One-off sounds that overlap speech, a door slam, a cough, another voice, are not separable by this class of model and will largely remain.
Which model size should I pick?
Balanced is the default for a reason: it audibly out-suppresses Fast at a small time cost. Choose Fast when the clip is long or the device is modest, and Ultra when the noise is heavy and you can wait. It runs the largest model and takes the longest.
Why does the result come out as WAV?
The models work at 48 kHz and the tool writes their output without a further lossy pass. Encoding back to MP3 would discard part of what was just cleaned; WAV keeps it, at the cost of a larger file.
Does my recording leave my device?
Supported file contents are processed locally in your browser through the site's neural runtime. The model files are fetched once and cached; the audio itself is not part of any upload.
Why does the effect strength slider exist?
Full suppression can feel sterile on some material. Breaths and natural room tone are part of how a voice reads. Pulling the slider back mixes a measured amount of the original recording into the cleaned one; the two are time-aligned, so the blend stays phase-coherent.
A soloed instrument or music came out damaged. Why?
The models are trained on speech. They treat sustained tonal content that does not look like a voice as noise, so music, tones and some sound effects can be attenuated. Use it on spoken recordings: interviews, voiceovers, meetings, lectures.