Remove fan noise, hiss, hum and other background sound from a recording via AI noise suppression - locally in the browser, no upload at all.
Running locally on your device ...
Running locally on your device ...
0%
Your files never left your device
No. Everything runs in your browser - your file never leaves your device. How this is verifiable
A voice memo with fan noise in the background, a video interview with street traffic, a podcast take with room hum and a computer fan - constant background noise is the most common reason an otherwise usable recording sounds unprofessional. This tool runs a small neural network (RNNoise, open source and battle-tested for years in video-conferencing software like Jitsi/Mozilla) that tells apart such stationary background noise from speech and subtracts it.
Technically, RNNoise processes the recording in tiny 10-millisecond chunks at 48 kHz - for each chunk, the network estimates how much of it is speech versus noise, and only dampens the latter. On real, steady background noise (fans, hiss, mains hum) the effect is often dramatic; on short, unpredictable sounds (a single knock, a door slam) the method naturally does less, because it is specifically trained to recognise STATIONARY patterns. Multi-channel material (stereo) is denoised per channel independently, so the stereo width is preserved.
The entire process - from the neural network to the final encode - runs as WebAssembly entirely in the browser, without cross-origin isolation or shared memory (hence single-threaded rather than multi-threaded, but still fast enough for everyday use). No server ever sees the recording.
| Input formats | MP3, WAV, M4A, AAC, OGG, OPUS, FLAC |
|---|---|
| Auto-conversion | MP4 |
| Output formats | WAV, M4A |
| Batch processing | No |
| Processing | Locally in your browser (WebAssembly) |
| File upload | None |
Limitations: Works best against CONSTANT background noise (fans, hiss, hum) - short, irregular sounds (knocks, clicks, thunder) are barely suppressed; this is not a general repair for damaged recordings. With very loud noise relative to the voice, the AI can leave audible artifacts. Processing is single-threaded (no cross-origin-isolation mode on this site) - a little more patience needed for very long files than shorter ones. Honest per-device length cap applies as for every audio tool.
No, noise removal runs entirely locally in the browser via WebAssembly.
Constant background sound like fans, hiss or mains hum - short, irregular sounds barely.
RNNoise, a small recurrent neural network that has been used in production video-conferencing software for years - here running as WebAssembly in the browser.
Yes, each channel is processed independently - left and right stay separate.
Compress audio · Reverse audio · Mono to stereo · Create a spectrogram