Noise remover
Drop a recording. The classifier picks the right method, shows why, and lets you save the clean file.
Nothing is uploaded. Everything runs in this tab with @audio/denoise.
Noise removal that shows its work
Cloud noise removers hide what they do and charge per minute. This page runs @audio/denoise, thirteen open-source restoration methods plus a classifier, in your browser. Drop a file, read which method was chosen and why, listen, save. Nothing leaves your device.
Background noise, hum, clicks, wind, plosives, reverb
- Hiss and fans: Wiener / MMSE-LSA gain with decision-directed SNR tracking, or OM-LSA when speech is present.
- Mains hum: adaptive notch bank on 50 or 60 Hz and harmonics, tracking drift.
- Clicks, pops, crackle: detection and LPC interpolation, the same family as vinyl restoration tools.
- Clipping: auto-detected clip level, waveform reconstruction.
- Wind, plosives, sibilance, reverb: dedicated dewind, deplosive, de-esser and dereverb stages.
When to pick a method by hand
Automatic mode handles the common cases. If you know the problem, choose the method: dehum for a ground loop, declick for a scratched record, declip for a hot recording, gate for silence between takes. The output plays in the page so you can compare before saving.
Questions
How does automatic mode choose? A classifier measures the recording: hum energy at 50/60 Hz and harmonics, click density, low-frequency rumble, noise-floor stationarity and sibilance. It picks the specialised method that scored highest and shows the plan.
Which noises can it remove? Steady background noise (fans, hiss, air conditioning), mains hum, clicks and pops, vinyl crackle, clipping, wind rumble, microphone plosives, sibilance, room reverb and noise between phrases.
Is this AI noise removal? No. Every method is classical signal processing: Wiener and OM-LSA spectral gain, spectral subtraction, adaptive notch filters, LPC click interpolation. It runs instantly on any device and never sends audio anywhere. For heavy speech-over-noise, ML models can do better; for hum, clicks, wind and hiss, these are the same algorithms the professional tools use.
Does it work on music? Yes for hum, clicks, crackle and clipping. Broadband denoising on music is gentler than on speech; use it lightly and compare with the original in the player.