Remove background noise from audio
Reduce the noise in meeting, lecture and interview recordings or in the audio of streams and videos so voices are easier to hear. The open-source speech enhancement AI “DeepFilterNet3” runs inside your browser, so the audio is never sent anywhere. No app to install. You can load the cleaned audio straight into Transcriber, a transcription tool by the same author.
The first run downloads an AI model (about 23 MB); later runs use the copy stored in your browser. Your files are never uploaded.
Settings
How to use
- 1
Add audio or video
Drop audio such as MP3, WAV, M4A or FLAC, or video such as MP4 or MOV. You can process several files at once.
- 2
Choose the mode and strength
The default is “Reduce noise” at “Standard”. Choose “Strong” if noise remains, or “Light” if the voice itself sounds cut back. To also fix room echo, clipping or muffled phone-like sound, set “Mode” to “Restore the voice” (for PCs that support WebGPU). To process a stereo recording faster, turn on “Save as mono”.
- 3
Run and save
The first run downloads the AI model (about 23 MB). “Restore the voice” downloads a separate AI model (about 39 MB) the first time. The output format can match the original file or be chosen from MP3, WAV and others.
How noise removal works and choosing the strength
DeepFilterNet3 is an AI that splits the audio into frequency components every 10 milliseconds, keeps the components of the voice and suppresses the components of the noise. It suppresses steady sounds such as an air conditioner or a fan, and also changing sounds such as outdoor noise. Music and singing can be cut like noise, so it is not suited to songs.
“Strength” is the upper limit on how much the noise is suppressed. “Strong” has no limit, “Standard” goes up to 24 dB and “Light” up to 12 dB; the lighter the setting, the more of the original sound remains. The default is “Standard”. Processing runs on the CPU: on a PC we tested (Ryzen 7 8745H), an hour-long mono recording took about 9 minutes and a stereo one about 25 minutes. The AI model (about 23 MB) is downloaded only the first time and stored on your device for later use.
Setting “Mode” to “Restore the voice” runs RE-USE, a speech restoration AI released by NVIDIA, in the browser. Besides noise, it also fixes room echo (reverberation), clipping and muffled sound such as phone audio. Steady noise such as an air conditioner is reduced better by “Reduce noise”. It runs on the GPU (WebGPU); on the PC we tested (the integrated GPU of a Ryzen 7 8745H), one minute of 48 kHz audio took about 2 minutes. On smartphones there may not be enough memory, and processing can stop or the page can reload.
Specifications of this tool
- Input
- MP3, WAV, M4A, AAC, OGG, OPUS, FLAC, WEBM, MP4, MOV, M4V, MKV, AIFF, WMA, M4B, M4R, MKA, 3GP, MP2
- Output
- Same as input, MP3, M4A, WAV, OGG, FLAC, AAC, OPUS, AIFF, M4R
- Limits
- No limit on the number of files (batch processing). No file size limit (processed within your device's memory). Output up to 1 GB per file (2 GB for AIFF)
- Free, no sign-up, no app to install
- Files are never uploaded (processed in your browser)
Frequently asked questions
Is my audio sent to a server?
No. Noise removal runs in the browser on your device and the audio data is never sent anywhere. The AI model (about 23 MB) is downloaded from the internet only the first time. No sign-up is needed. Processing statistics used to fix bugs (device specs, error types and so on; never the audio or its file name) are sent to this site's server, and you can turn this off with the switch under “Help us improve the service” at the bottom of the page.
What kinds of noise can it remove?
It reduces steady sounds such as air conditioners, fans and computer hum, as well as outdoor noise. Music and singing can be cut like noise. To pull only the vocals out of a song, use this site's “Acapella extractor” instead. For room echo (reverberation) or clipping, set “Mode” to “Restore the voice” (for PCs that support WebGPU).
Can it remove noise from the audio of a video?
Yes. Dropping a video produces an audio file with the noise reduced in its audio track. It cannot write a video with the picture.
Can it handle long recordings?
Yes. The audio is read and processed a little at a time, so recordings several hours long work too. Processing time grows roughly in proportion to the length; on a PC we tested (Ryzen 7 8745H), an hour-long mono recording took about 9 minutes. To process a stereo recording faster, turn on “Save as mono”. Keep this page open while it is processing.
Help us improve the service
If a conversion fails, sending us the circumstances helps us track down the cause. Knowing which devices and formats fail most often, and how long processing takes, lets us revise the supported formats and fix the bugs behind them. If you don't mind, leaving this on is a real help.
What is sent
- The tool you used and the settings you chose (output format, quality, etc.)
- Your device specs (GPU, memory, core count)
- The input format with a rough size
- How many files succeeded or failed and how long it took
- The error type and details when something fails
What is never sent
- The files themselves and file names
- IP addresses
The data goes only to this site's own server, never to a third-party service.
For the full list, including the details each tool sends, see “Processing statistics” in the privacy policy.