Podcasts and interviews
Try reducing fan noise or a steady room background behind an audible speaker. Check both speakers if their distance from the microphone differs.
Local AI audio tool
Remove background noise from voice recordings for free
Reduce background noise in voice recordings for free. Preview, compare and download a WAV file with local AI in your browser, without uploading audio.
Choose an audio recording
Best for spoken audio
RNNoise processes voice recordings in short frames. It can reduce steady noise such as fan or room noise while keeping speech understandable.
It cannot reliably remove another speaker, strong echoes, or noise that completely masks words. Music may change or sound unnatural. Listen to the preview before processing the full recording. Maximum length: 30 minutes.
Audio processing and temporary results stay on your device. Only the tool code is downloaded.
RNNoise · Open-source project
How to remove background noise from audio
Start with a short comparison before processing a whole recording. Keep the original file so you can choose the version that sounds most natural.
Choose a voice recording
Select an audio file from your device. Mono and stereo recordings up to 30 minutes and 500 MB are supported. Decoding depends on the format and your browser.
Preview the first 30 seconds
Run the preview, then switch between Original and Denoised at the same point. Listen to quiet words, breaths and pauses as well as the noise itself.
Adjust the blend
The default 80% blends processed audio with some original sound. Lower it if speech sounds thin or metallic. A higher percentage uses more of the denoised signal; it is not a promise of better quality.
Process and save the recording
Choose Full recording and run again when the preview sounds right. Download the WAV before leaving, or send the result to transcription, trimming or format conversion.
Which recordings benefit from noise removal?
Voiceovers and lessons
Clean up narration recorded at a desk or in a home office. Listen for softened consonants and word endings before using the result in an edit.
Voice notes and meeting recordings
Reduce distracting background sound in spoken notes. Overlapping voices remain voices: this tool cannot reliably select one person from a conversation.
Get a cleaner result without losing natural speech
Judge speech, not only silence
A quieter pause does not necessarily mean a better recording. Compare intelligibility and tone during speech, ideally at a similar listening volume.
Use a lighter blend when needed
If the voice sounds watery, gated or metallic, lower the denoised percentage and run again. Repeatedly processing the output can compound artifacts; return to the original instead.
Know what cannot be recovered
Strong wind hitting the microphone, clipping, heavy reverberation and words buried in noise may remain damaged. Recording closer to the speaker in a quieter space is preferable when a retake is possible.
Audio noise removal FAQ
Is my recording uploaded to a server?
The denoising runs in your browser using RNNoise. The page downloads its code, but the recording and temporary WAV result stay on your device. Download the result before clearing browser storage or leaving the workflow.
Which audio formats can I use?
You can select WAV, MP3, M4A, AAC, FLAC, OGG and Opus files, but selection does not guarantee decoding support. The actual codec and browser determine whether the file opens. The tool checks the recording before processing.
What is the exported format?
The result is a 48 kHz, 16-bit PCM WAV file with the input channel count preserved for mono or stereo. WAV can be much larger than a compressed source. Use Audio Converter afterward if you need another format.
Can this remove music or isolate a singer?
RNNoise is intended for speech noise reduction. Background music may be changed unpredictably, and singing may lose detail. Use Vocal Separator when your goal is to separate vocals and accompaniment.
Will denoising improve transcription accuracy?
It may help when steady noise interferes with otherwise clear speech, but it can also remove useful speech detail. Compare the original and cleaned recordings, then review the transcript rather than assuming the cleaned version is more accurate.
Why is processing slow or failing?
Speed depends on recording length, channel count and your device. Try the preview first, keep the tab open and make sure browser storage is available. An unsupported codec or insufficient storage can prevent processing even when the filename looks valid.
Continue editing your audio
Audio to Text
Turn the cleaned recording into editable text and subtitles. Review names and quiet words against the audio.
Vocal Separator
Separate vocals from accompaniment when music, rather than room noise, is the main problem.
Audio Volume Booster
Adjust a quiet recording after checking the noise. Raising volume also makes any remaining noise louder.
Reproducible checks
Verify Audio Noise Remover with known inputs
Generated test patterns, tones and synthetic speech, with source files and a reference transcript. Use the small fixtures to check a specific operation before processing your own media.
- Reproduce
- Process the noisy-tone fixture and compare it with the clean-tone reference at matched playback volume. Save the output for inspection.
- Inspect the saved result
- Check the exported file, duration and whether the tone or added noise changed. Keep the selected denoising settings in your notes.
- Counterexample and limit
- This is a negative/control signal, not noisy speech. A speech-oriented model may attenuate the tone; no speech-quality improvement is claimed.
What has actually been verified
The record below is code-level verification in Node.js, not a complete browser UI or codec-compatibility test. Parser checks use the website’s functions; fixture checks describe inputs only. No AI quality, browser timing or large-file performance score is claimed.
2026-10-27 16:38:27 UTC · Node.js v24.19.0 · win32/x64
WAV fixture properties · Passed
Input-file verification only. It does not certify conversion, AI inference or output quality.
- Duration (s)
- 2
- Sample rate (Hz)
- 44100
- Channels
- 1
- Bit depth
- 16
- Bytes
- 176444
- Peak amplitude
- 0.25
For your own browser test, record browser/version, operating system, output settings and whether decoding, processing and saving all succeed. A reference result is not a promise for another device.