Clean up a voice recording
A recording made in a noisy room is tiring to listen to and hard to transcribe. Drop it here and a speech-enhancement model removes the background while keeping the voice — on your own device, so an interview or a confidential meeting never leaves it.
- Runs entirely in your browser
- Works offline
- No upload
- No sign-up
- No watermark
Loading the tool…
How it works
- Drop an audio or video file: a voice memo, an interview, a meeting or a lecture.
- Choose the strength and click “Clean up the recording”. The first time, the model (8.6 MB) and the engine that runs it are downloaded and kept in the browser.
- Compare the original with the cleaned version, adjust the strength if you like — the result updates at once — and download WAV or MP3.
Why nothing is uploaded
Every operation on this page is done by code running inside your browser tab, using the same engine that renders web pages. The file is read from disk into your tab’s memory, transformed there, and written back out as a download. It is never sent anywhere — not to us, not to a third party.
Verify it yourself
- Open your browser’s developer tools (F12) and select the Network tab.
- Load your file and run the tool.
- The only requests you will see fetch the tool’s own code — and, for a few heavy tools, their open-source engine (also from this site) or an AI model from Hugging Face — plus one small page-view ping to loreatec.jp (page address and title, nothing more). None of them carry your file.
Frequently asked questions
What kind of noise does it remove — and what can it not do?
It is trained to keep speech and remove everything else: steady noise such as hum, hiss, fans, air conditioning and road noise, and changing noise such as keyboards, clatter and the chatter of a café. What it cannot do: separate one voice from several people talking over each other (it keeps all the speech), remove music playing under a voice, or undo strong echo in a large bare room. Nor can it repair distortion from a recording that was too loud (clipping) or make a distant voice sound close — those are different problems.
Which strength should I use?
The strength is the most noise the tool may remove. The default, 24 dB, takes out most of it and leaves a faint trace of the room, which sounds natural. Lower values keep more of the room sound and are gentler on the voice; “Maximum” removes all the noise it can, which can make the pauses sound unnaturally dead. You can change it after cleaning and the result updates at once, without processing again.
Is my recording uploaded anywhere?
No. The audio is decoded and cleaned inside this tab. The only downloads are the model itself — DeepFilterNet 3 by Hendrik Schröter and contributors, 8.6 MB, MIT licence, fetched from Hugging Face — and the engine that runs it (27 MB, from this site). Both are kept in the browser, so after the first time the tool works offline too.
Can I use a video, or a stereo file?
Yes. Drop an MP4, MOV or WebM and its sound track is cleaned. Most audio formats open directly; unusual ones such as WMA or the AMR files of old phones are opened by an audio engine that downloads on first use (about 32 MB). Stereo and multichannel files are mixed into one channel first, so the result is mono, which is what speech needs. To put the cleaned sound back into the video, use “Change a video’s audio track” in this section.
How long can the recording be, and how long does it take?
The audio is cleaned in 10-second blocks, and the model’s memory is carried from one block to the next, so a long recording comes out exactly as if it had been processed in one go. On a recent laptop it takes about a tenth of the recording’s length — roughly a minute for a ten-minute interview. The real limit is memory: the whole recording has to fit in the tab (about 400 MB for ten minutes), so split anything longer than an hour or so first.
Why does the cleaned voice sound slightly different?
The model rebuilds the sound from what it judges to be speech, so it also trims some room reverberation and breath noise, and the level can come out a little lower. On a recording with no noise it changes little. Listen to both versions before you use the result, especially with heavy noise, where any noise reduction can leave the voice sounding thin or watery.