Remove background noise without uploading

What leaves your device, action by action

Only the two Enhanced actions send sound to our servers, and neither happens unless you click it.

Check it yourself

You don’t have to take our word for it. In Chrome, Edge or Firefox, open the developer tools (F12, or Cmd+Option+I on a Mac) and go to the Network tab. Clear the list, pick your file above, play the preview and click Clean full file.

You will see the page download the noise removal model, about 16 MB the first time, and some code. Switch the filter to Fetch/XHR to see what your browser sends. You will find only small requests for your sign-in state and page statistics, plus the billing record from the table for files over the free length. None of them is anywhere near the size of your file. Your file appears only as a blob: address, which is your browser reading it locally, not a request to any server.

What runs in your browser

Standard is DeepFilterNet 3, an open speech enhancement model, running through ONNX Runtime in WebAssembly. The model file is downloaded from our site, then it does all the work on your computer or phone: decoding the file, removing the noise and writing the result. For a video, only the soundtrack is replaced and the picture is copied as it is.

Because the work happens on your device, speed depends on it. A one-hour recording takes longer on an older laptop than a cloud tool would. If your browser can’t read a file format on your device, the page tells you, and only Enhanced is offered for that file.

Who this is for

Interviews with sources who were promised confidentiality, therapy, medical or legal notes, internal meetings, user research sessions and unreleased episodes: recordings you may not be allowed to upload to a cloud service. Standard lets you clean them without creating an account or handing the file to anyone.

We don’t claim that using Standard makes you compliant with a particular regulation or company policy. What we can say is that the file isn’t transferred, and you can confirm that yourself. Check your own rules for what that means in your case.

The trade-off: Standard is smaller than Enhanced

A model small enough to run in a browser can’t do everything a larger server model can. In our tests Standard handled traffic, chatter, hum and wind well, but Enhanced removed far more steady hiss and kept voices brighter on some recordings. The café sample above has both versions so you can compare.

If a recording is private and Standard isn’t enough, that is your decision to make. Preview Enhanced sends only 30 seconds of sound, so you can hear the difference before uploading the whole file.

On a shared computer

So that a page refresh doesn’t lose your work, the page keeps your file and its results in this browser’s own storage for 24 hours, and at most your last 5 files. They stay on this device and are not uploaded. On a shared or work computer, download your result, then clear this site’s data in your browser settings, or use a private window, which discards everything when you close it.

Common questions