Audio Noise Reducer: clean up any recording

An audio noise reducer is a tool that removes unwanted background sounds from a recording. It works by analyzing the audio frequencies and suppressing the steady noise (like hiss, hum, or fan noise) while preserving the main audio (like human speech or music).

No. The denoiser runs entirely in your browser using FFmpeg compiled to WebAssembly. Your file is read from disk, processed in memory and re-saved to disk. No part of the audio, no part of the filename and no part of the waveform is sent anywhere.

Steady background noise: hiss, hum, air conditioning, computer fans, electrical buzz, tape hiss and traffic rumble. It is less effective on intermittent sounds like a single cough, a door slam or a dog bark, because those do not look like steady noise to the FFT algorithm.

MP3, WAV, M4A, AAC, OGG, Opus, FLAC and WebM. If your file is in a less common format, convert it to WAV or MP3 first and then run it through this tool.

MP3 at 192 kbps. That is fine for voice and podcast content and plays in every browser, phone and player. If you need lossless output, this tool does not offer it yet.

The first time you process a file, the tool downloads the ffmpeg engine (around 25 MB of WebAssembly). After that it is cached by your browser and subsequent runs start immediately. If you have already used other tools on ToolDeck that use ffmpeg, the cache is shared.

No. The denoiser is tuned to suppress steady noise, not to isolate or remove voices. If you turn up Noise reduction too far on a quiet voice recording, the voice itself can start to sound hollow or metallic. Stick to Medium or Light unless you really need more.

Not directly. This tool targets background noise, not music separation. If you need to strip music from a video, use a dedicated stem separation tool.

No. The audio is not time-stretched, pitch-shifted or otherwise retimed. The output has the same duration and the same pitch as the input.

Yes. Play the original with the built-in player, then play the cleaned file. Both players are in the tool. If the result is not good enough, adjust the preset and try again.

The tool works in modern mobile browsers but processing is much slower than on desktop. Files under a few minutes work. For longer files or faster turnaround, use a laptop or desktop.

Noise floor is the level, in decibels, below which audio is treated as noise. Anything quieter is suppressed. A higher value (closer to 0) is more aggressive. The presets set sensible values: -50 dB for light, -40 dB for medium, -30 dB for strong.

Noise reduction is how much attenuation is applied to the identified noise, in decibels. Higher values remove more noise but risk making the voice sound thin. The presets use 6, 12 and 20 dB.

No hard limit, but processing time scales with file length. A 5-minute recording is usually under a minute on a modern laptop. Hour-long files can take several minutes and use a lot of memory, so split them first if you can.

Yes. There are no restrictions on the cleaned file. Whatever rights you hold over the original recording, you keep over the result. No ToolDeck branding, no watermark, no attribution required.

No. The output MP3 has no embedded ToolDeck tag, no ID3 comment added by us and no audible watermark. The file is exactly what the encoder produced from your input.

If you set Noise reduction or Noise floor too aggressive, the tool starts removing the voice as well as the noise, which makes it sound hollow. Reset to a lighter preset and try again. For very quiet voice recordings, use Light and expect only modest cleanup.