Clean the hum, hiss, fan, and traffic out of a voice recording. An AI noise model does the work on your own device, you hear the before and after, and the download is yours in WAV or MP3.
Every phone memo, lecture recording, interview, and late-night podcast take has the same enemy: the room. The air conditioner you stopped hearing years ago, the laptop fan, the refrigerator two rooms away, the traffic outside. This tool runs a neural noise-suppression model over your recording, keeps the voice, and drops the background, entirely on your own device.
The model is RNNoise, a small recurrent neural network developed at Mozilla and Xiph.Org that became a quiet industry standard: versions of it run inside major voice-chat apps to clean microphones in real time. Here it runs as WebAssembly inside the page, about 110 KB, which is why the tool loads instantly and works offline once open.
Neural noise suppression is at its best on steady, continuous noise: hum, hiss, fans, HVAC, traffic rumble, and general room tone under a clear voice. That covers the great majority of "my recording sounds bad" cases.
It is not magic. Sudden one-off sounds (a door slam, a cough, cutlery), music mixed with the voice, and a second conversation in the background overlap the speech itself, and no small model can cleanly pull those apart. Echo and heavy reverb are also a different problem from noise. If a recording is mostly noise with a faint voice underneath, expect an improvement, not a studio take.
The page loads the model to your browser; your recording is decoded, processed, and re-encoded in your device's memory and never transmitted. There is no server-side audio service, no queue, and nothing to delete afterwards, because nothing was ever received. Load the page, go offline, and it still works.
Drop the audio file into the tool, click Remove noise, and an AI noise-suppression model separates the voice from the background on your own device. You can hold a button to compare against the original, dial the strength up or down, and download the cleaned file as WAV or MP3. No account and no upload.
Steady, continuous noise is its specialty: air conditioning, computer fans, electrical hum, hiss, traffic rumble, and room tone. Sudden one-off sounds like a door slam, a cough, or someone talking in the background are much harder, since they overlap the voice in ways a noise model cannot cleanly separate.
At full strength on a very noisy recording you can sometimes hear a slightly processed, underwater quality. That is why the tool has a strength slider: it blends the cleaned audio with the original, so you can back off to 70 or 80 percent and keep the voice natural while most of the noise stays gone.
No. The noise-suppression model (RNNoise, about 110 KB of WebAssembly) loads into the page and processes your audio on your own device. Interviews, meetings, voice memos, and lecture recordings never leave your browser, and the tool keeps working offline once loaded.
It accepts anything your browser can decode, which covers MP3, WAV, M4A/AAC, OGG, and usually FLAC. Export is lossless WAV or a 192 kbps MP3. If your audio is stuck inside a video, extract it first with the Audio Converter on this site, which is also browser-only.
It depends what you need. Adobe's enhancer does heavy studio-style reconstruction on their servers, which means uploading your audio and an account. This tool does honest noise suppression on your device: less dramatic on terrible recordings, but private, instant, free, and with no file length meter.