Use this background noise remover to reduce unwanted sound in a voice recording. Choose a local audio file, compare the cleaned result with the original and download MP3 or WAV.
Reduce background noise in speech
Choose a voice recording up to 25 MB and 3 minutes. Compare the original with a cleaned mono MP3 or WAV. Audio processing stays in this browser.
Original audio
Output is mono: stereo input is averaged to one channel. This is for speech; music, echo and other voices may remain or change.
Compare at the same device volume. Listen for lost words or muffled speech before using the result.
Download cleaned audioOpen-source notices
RNNoise 0.1 model via @jitsi/rnnoise-wasm 0.2.1. RNNoise: BSD-3-Clause; Jitsi wrapper: Apache-2.0 (ES module export removed for worker loading). RNNoise source · WASM wrapper source. MP3: lamejs 1.2.1 / LAME, LGPL, unmodified library. Encoder source. Full license notices retained in worker source.
How to remove background noise from audio
- Choose a playable voice recording up to 25 MB and 3 minutes.
- Start with Balanced. It mixes 80 percent processed audio with 20 percent original audio.
- Choose MP3 or WAV and create the cleaned file.
- Listen to the original and result at the same device volume. Check complete words and consonants, not just quieter pauses.
- If speech sounds muffled, try Gentle. If too much noise remains, try Full and compare again.
- Download the version that sounds most useful. Your original file stays unchanged.
Which noises can this reduce?
The local RNNoise engine is designed for speech noise suppression. It can help with some fan-like noise, hiss and general background noise, depending on the recording. A spoken note recorded near a fan is a reasonable file to try; the improvement will vary with the distance, noise level and microphone.
This is not guaranteed removal of every unwanted sound. Wind, sudden impacts, overlapping speakers and loud music can be difficult. It does not isolate a named speaker, separate singing from instruments or reliably remove room echo. Missing or already clipped speech cannot be reconstructed.
Gentle, Balanced or Full?
The options control how much processed audio is blended with the original. Gentle uses half processed and half original audio; Balanced uses 80 percent processed; Full uses only the processed signal. They do not represent a measured percentage of noise removed or different trained models.
Keeping some original sound can reduce an overly processed feeling, but brings back some noise as well. Full can sound cleaner in pauses while altering quiet words or voice texture. Always compare a section with speech before choosing it.
The model can take a short time to adapt at the beginning. Some noise can remain near the edges. The tool compensates for the engine’s short processing delay and exports the source’s decoded sample count, but filtering can still change transients and boundaries.
Mono output and file formats
This version creates one mono channel at 48 kHz. Stereo input is averaged before processing, so stereo separation is not preserved. If a stereo source contains opposite-phase signals, averaging can cancel some sound. Use an appropriate mono source when this matters.
MP3 output is genuinely encoded at 128 or 192 kbps. Re-encoding is lossy and may add a little padding. WAV contains 16-bit PCM and is generally larger. If processed samples exceed the output range, the file’s level is reduced before encoding. This is not loudness normalization or a guarantee against later codec overshoot.
MP3, WAV, M4A, OGG and FLAC input depend on what this browser can decode. The source may be resampled to 48 kHz. Metadata, cover artwork and extra tracks are not copied. An unsupported extension cannot be fixed just by renaming it.
Limits and privacy
One file, up to 25 MB and 3 minutes. Mono or stereo input only, with a decoded limit of 18 million channel samples. Decoded audio and the processing engine need memory; try a shorter recording on a slow device.
Noise reduction and encoding happen in this browser. The tool does not upload the audio contents to our server, call a remote AI API, save it in browser storage or request microphone permission. The website’s usual assets and analytics can load separately. Keep the page open until it finishes. Cancel stops the worker and discards the partial export.
This edits local audio files, not a live call, a streaming link or a video’s audio track. For a quieter result that only needs more volume, use the audio volume booster after checking the cleaned voice.
Common problems
Speech sounds muffled: use Gentle or keep the original. Stronger processing is not always better.
Other people are still audible: this is noise suppression, not speaker separation. It may preserve speech you wanted to remove.
Music sounds odd: the tool is intended for speech. Avoid using it as a music restoration or vocal isolation tool.
File will not load: try a playable MP3 or WAV in an up-to-date browser with Web Audio, WebAssembly and worker support. Very short or damaged files may fail.
Frequently asked questions
Can I remove background noise from a voice recording online?
You can try this local speech noise suppressor, compare the result and download it. Results depend on the noise and source quality.
Will it remove all noise?
No. Some noise may remain, and processing can change the voice. Listen before keeping the result.
Can it reduce fan noise or hiss?
It can help with some recordings containing these noises. It cannot promise the same improvement for every microphone and noise level.
What does 80 percent processed mean?
Balanced mixes 80 percent cleaned signal with 20 percent original signal. It does not mean 80 percent of the noise was removed.
Can it remove echo, music or another speaker?
Those tasks are not reliably supported. This version is for general speech noise suppression.
Will the download stay stereo?
No. Stereo input is averaged into one mono channel, and the cleaned download is mono at 48 kHz.
Is the audio uploaded or stored?
The tool processes it locally and creates a separate download. It does not upload its contents or store the selected file in browser storage.
Related tools
Audio volume booster · MP3 cutter and ringtone maker · Video cutter · All tools