Free Background Noise Remover
Remove background noise, wind, traffic, hiss and hum from voice recordings with an AI model that runs on your device. Nothing is uploaded and no account is needed.
Many AI audio cleaners upload your recording and need an account — here the AI model runs on your device, with no account.
Strip background noise out of a recording (traffic, wind, chatter, fans, hiss, hum) with an AI model that runs in your browser. Hear the before and after before you commit.
Drop files here, or click to browse
MP3, WAV, M4A, AAC, OGG, FLAC, WEBM, MP4, MOV, MKV, M4V · processed on your device
or press Ctrl+V to paste from your clipboard
You can drop multiple files at once
Your files never leave your device. All processing happens locally in your browser.
🔒 Your recording is processed entirely in your browser — nothing is uploaded.
Common uses
- Remove Background Noise from PodcastsClean hiss, hum and fan noise from podcast recordings in your browser, with a before/after preview.
How It Works
The default Voice (AI) mode runs DeepFilterNet3, an open-source speech-enhancement neural network (MIT/Apache-2.0), on your own device through ONNX Runtime in a background worker. Its 8 MB model and 14 MB AI engine are downloaded once from this site and cached; your recording is never uploaded. ffmpeg.wasm (about 31 MB), also served from this site and cached, reads your file and writes the result.
- 1Pick the kind of noiseVoice (AI) is for speech over traffic, wind, chatter, fans or hum: the network estimates, 100 times a second, how much of each frequency band is voice and filters the low frequencies with a short learned filter, so it follows a voice through a changing room. It treats music as noise. Hiss & hum runs afftdn, an FFT denoiser that subtracts a steady noise floor and is safe on music; Learn the noise teaches that filter a moment where only the noise is heard.
- 2Strength sets how much background is leftFor Voice (AI), strength is an attenuation limit: at 100% the model removes all the noise it finds, at Medium (80%) it lets the background drop by at most 24 dB, at Light (50%) by 12 dB, so some room tone stays and voices sound natural. For Hiss & hum it sets the filter's noise reduction and assumed noise floor.
- 3A 15-second A/B preview runs the identical processingPress Preview and 15 seconds of your recording go through exactly the same model or filter as the full file, then the cleaned and original takes play through one AudioContext so you can switch between them without losing your place. With Match original loudness on, the cleaned take is brought to the original's level, so you compare the noise rather than the volume.
- 4Long files in pieces, with a Cancel that really cancelsffmpeg decodes the file once into 30-second pieces at 48 kHz; each piece goes through the model in a background worker and is freed as soon as it is read, so memory stays bounded and the progress bar moves. The cleaned audio is then written as WAV or MP3 at your recording's own sample rate, or put back under the untouched picture of a video. Cancel stops both workers at once.
Your recording never leaves the device. The model, the ONNX engine and the audio engine are served from this site's own servers (or an identical official copy if those can't be reached) and cached by your browser; the file itself is only read inside your browser.
Guide
Worked example
A 10-minute stereo voice recording at 44.1 kHz, made in a room with an air conditioner running, is about 106 MB as WAV or about 14 MB as MP3 (this tool writes MP3 at 192 kbps). Pick "Voice (AI)", leave Strength at Medium, press "Preview 15 seconds" and flip between "Cleaned" and "Original". If the voice still sounds natural, press "Remove noise" to process the whole file. On a computer the AI mode runs several times faster than real time, so expect a few minutes at most.
Which settings to use, and the limits
- Match the mode to the noise. "Voice (AI)" is built for speech in traffic, wind, chatter and fans, and it treats music as noise. For a song with tape hiss or mains hum use "Hiss & hum". For a steady noise under speech or music, such as a fan, use "Learn the noise" and point it at a second or two where only the noise is playing.
- Start at Medium and listen. Strength runs from 5 to 100, with Light at 50, Medium at 80 (the default) and Strong at 100. Higher removes more noise but can make voices sound thin or underwater. The preview takes 15 seconds from the middle of the recording and uses exactly the same processing as the full file.
- Turn on "Match original loudness" when comparing. Cleaning lowers the level, so the original always sounds a little louder in the preview and your ear prefers it. With this option on, the cleaned take is brought back to the original's average loudness with a limiter, so you hear the difference in noise and not in volume.
- Choose WAV for editing, MP3 for sharing. WAV keeps full quality and the source sample rate. MP3 is encoded at 192 kbps and is much smaller. For a video file, "Video" keeps the picture untouched and replaces only the sound; MP3 or WAV give you just the audio.
- Know the file size limit. The largest file accepted depends on the device: about 2 GB on a normal computer, about 1 GB on a computer reporting 4 GB of memory or less, and about 400 MB on a phone. Long files are cleaned in 30-second pieces, so a 30-minute recording is fine on a phone, but expect roughly half the recording's length or more. Large videos skip the preview to keep the browser responsive.
- Expect a first-time download. The audio engine that reads and writes your file is about 31 MB, and the AI voice mode adds about 22 MB (model and AI engine). Both are cached afterwards. Your recording is not uploaded.
Common problems and how to fix them
- It says the audio file could not be cleaned up.
- The file may be damaged or in a format the browser engine cannot read. Try again with the original recording, or save it as MP3 or WAV first and re-open that. If the file is very large, cut a shorter part out with the trimmer and process that.
- The file is rejected as too large.
- The message shows your file size and the limit for this device. Trim the recording to a shorter piece, or try on a desktop computer, which allows larger files than a phone does. The cleaned parts can be joined afterwards.
- The AI voice model would not load.
- The model and AI engine come from this site and need a working connection the first time. Check your connection and press the button again. If it keeps failing, choose "Hiss & hum", which needs no voice-model download.
- The video has no audio track to clean.
- Some screen recordings and exports contain only picture. Check that the original plays with sound. If the sound is in a separate file, clean that file instead.
- It says the file changed or moved after I picked it.
- The browser can only read a file as long as it stays where it was when you chose it. This happens if you re-run a long job after moving, renaming or syncing the file. Choose the file again.
How it compares
The three modes suit different noise. The AI voice model is the strongest on speech in changing noise such as traffic or chatter, but it thins out music. Hiss & hum is a classic filter for steady noise and is safe on any material, but it will not remove a barking dog or a passing car. Learn the noise sits between the two: it removes one steady noise very precisely once it has heard a sample of it.
Related Tools
Video Noise Remover
Remove background noise from video in your browser: wind, traffic, fans and hum. The picture stays untouched and only the sound is cleaned. Free, nothing is uploaded.
Audio Silence Remover
Automatically detect and cut silent gaps from your audio with an adjustable threshold — entirely in your browser.
Audio Mixer — Add Background Music to Audio
Layer up to 8 audio files into one: put background music under a voice recording or podcast, with its own volume, start time, fade in and out, and loop. The music can dip automatically while the voice talks. Save as MP3, M4A or WAV. Runs in your browser; nothing is uploaded.