Local model processing
The RNNoise model is loaded only when you start processing. The source waveform stays in the browser instead of going through a cloud audio API.
Noise Reduction Online
Clean steady background noise from a short recording, compare the original and cleaned signal, adjust the mix, and download a local WAV.
Browser modelBefore and after previewNo credits required
Noise Reduction capabilities at a glance
Understand the model boundary before you process a recording.
- Input
- A browser-decodable audio file up to 50 MB and 5 minutes
- Output
- A locally processed 48 kHz mono WAV with adjustable wet/dry strength
- Model
- RNNoise WASM loaded on demand and executed in a Web Worker
- Privacy
- The source audio is decoded and processed in browser memory
A practical cleanup pass with an honest boundary
Noise Reduction is designed for steady room and recording noise, not as a promise to repair every damaged mix.
Compare before you export
Listen to the original and cleaned versions side by side so you can catch speech damage, musical artifacts, or an over-processed result before downloading.
Control the blend
Use the strength control to mix the cleaned signal with the original. A lighter blend can preserve texture when full suppression sounds too aggressive.
How to use Noise Reduction
The workflow keeps model loading, processing, review, and export visible.
- 01
Choose a short recording
Upload a browser-supported audio file under 50 MB and 5 minutes. Short voice clips, demos, and room recordings are a good first test.
- 02
Run local cleanup
The page resamples a mono working signal to the RNNoise frame rate and runs the model in a Worker so the interface can report progress.
- 03
Review and download
Compare the original and cleaned audio, adjust the strength when needed, then download the resulting 48 kHz mono WAV.
Where a noise reduction pass helps
Use it as a focused preparation step before a larger music or voice workflow.
Voice demos
Reduce steady fan, HVAC, or room noise in a short vocal idea before sharing a reference.
Sample preparation
Clean a noisy field recording or rehearsal fragment before trimming, sampling, or converting it to another format.
Transcription prep
Improve the starting signal for a later transcription or editing pass when the noise is steady and the speech or melody remains present.
Get a better cleanup result
Noise suppression is strongest when the unwanted sound is consistent and the wanted signal is clear.
Use steady noise as the target
RNNoise is a noise suppressor rather than a magic eraser. Constant hiss, fan, and room noise are more suitable than loud clicks, clipping, reverb tails, or another song playing underneath.
Listen for musical detail
Aggressive suppression can soften consonants, transients, or high-frequency detail. Compare both previews and reduce the blend when the cleaned version sounds watery or dull.
Keep the output boundary in mind
The current browser model exports a mono 48 kHz WAV. It is a cleanup reference and a useful next step, not a replacement for a full restoration suite or multitrack mix.
Noise Reduction questions, answered
Direct answers about the model, privacy, formats, and expected results.
Continue the audio workflow
Move from cleanup to the next concrete editing or analysis task.
Make a cleaner reference
Choose a short recording with steady background texture, preview the change, and export the version that serves your edit.
Reduce background noise