Audio Noise Remover (Background Noise Reduction)

Clean up a voice recording, interview, lecture or video soundtrack that has a steady background noise: hiss, a fan, an air conditioner, a computer, room tone or electrical hum. The tool measures the noise in every frequency band and turns those bands down where the noise is stronger than the sound you want, with an optional rumble filter and 50 or 60 Hz hum notches. It shows the background level before and after. Your files are processed in your browser and never uploaded.

How to use this tool

1. Choose one or more recordings.
2. Start with Strength 12 dB and the automatic noise profile.
3. If the recording starts or pauses with noise only, choose "Learn from a noise-only part" and enter that time for a more precise profile.
4. Add the rumble filter (80 Hz for voice) or hum removal if needed, press Calculate and compare.

Formula

For each frequency band: gain = 10^(−strength ÷ 20) where the noise dominates, rising towards 1 where the sound is well above the noise · noise floor = measured noise (dBFS) + 6 dB

This is spectral noise reduction with ffmpeg's afftdn filter, run by the ffmpeg engine (about 30 MB, downloaded once). It works from the measured noise alone and does not recognise voices or music, so the automatic profile needs pauses where only the background is heard. The noise level is measured first: automatically as the RMS of the quietest 10% of 50 ms moments in the first 5 minutes, or as the RMS of the noise-only part you give, whose spectrum the filter also learns. The filter's 25 ms processing delay is removed, so the result stays in sync with the original or a video. The rumble filter is a 12 dB-per-octave high-pass; hum removal places narrow notches at the mains frequency and its first five harmonics (for 50 Hz: 50, 100, 150, 200, 250, 300 Hz). The background level is the RMS of the quietest 10% of 50 ms moments.

Example

A voice recording with a steady hiss at −45 dBFS (about −50 dBFS in the speech band the results measure) gets a noise floor of −39 dBFS. With Strength 12 dB the background measured after processing is about −62 dBFS, 12 dB lower, while the loud parts of the voice change by less than 0.1 dB.