Voice noise cleaner
Reduce background noise in speech and compare the result before saving.
Files stay on your device. Nothing is uploaded.
Up to 25 MB and 3 minutes; mono/stereo, up to 48 kHz. Output: mono WAV.
About 1.3 MB of compressed model/runtime data loads on demand. Audio stays on your device.
For speech, not music mastering. Listen for lost syllables or artifacts; stereo is mixed to mono.
Related tools
Reduce background noise in a short spoken recording without sending it to an audio service. The local RNNoise model is designed for speech, so start with a voice note, narration or recorded conversation rather than music. After processing, listen to the original and cleaned version before downloading a WAV file. Noise reduction can also change the voice or remove quiet sounds; it cannot reliably separate overlapping speakers or restore clipped audio. Keep your original recording and choose the version that sounds better.
How to use
- Choose a supported audio or video file up to 25 MB and three minutes, with mono or stereo audio.
- Start noise reduction; the recording is mixed to mono and processed at 48 kHz on this device.
- Listen to the original and result, then download the cleaned WAV if the change suits your recording.
What it handles
- RNNoise speech denoising runs locally; the model and runtime load only after you start.
- Original and processed audio have separate labelled players, so you can judge the change yourself.
- Exports a 16-bit mono WAV at 48 kHz; stereo placement is not retained.
- Cancellation stops the worker, and Clear removes the working file and both preview URLs.
FAQ
Are my files sent to an AI service?
Voice cleaning decodes audio and mixes it to mono locally. Your file is processed in this browser, not uploaded to an inference service. Required models and runtime files are downloaded from this site after you start processing. The tool does not save your input or result in local storage. Keep the original file yourself and download any result you want to retain before closing the page.
Can it remove music, echo or another person speaking?
This is background-noise suppression for speech, not a stem separator or a repair service. Steady noise may be reduced, but music, reverberation and overlapping speakers can remain or sound distorted. Loud clipping cannot be undone. Compare the whole recording, including quiet syllables and pauses, and retain the original if the processed version sounds worse.
Why is the downloaded file WAV and sometimes larger?
The result is uncompressed 16-bit mono PCM at 48 kHz so it can be opened in common audio editors. A WAV can be larger than a compressed MP3 even when its background noise is lower. File compression and noise reduction are different tasks. Stereo input is mixed down, while the original recording remains unchanged on your device.
Updated