60%
Hear what people are saying
Background music is one of the most common reasons viewers struggle to follow speech, and it is hardest on people with hearing loss.
BBC TV audibility survey, 2011. RNID survey of members with hearing loss, 1998.
nomus takes the music out of any tab, your whole computer, or your phone, and keeps the voices. A small Rust engine does it on your device. Your audio never leaves it.
The real engine, running in this page
idle
60%
Background music is one of the most common reasons viewers struggle to follow speech, and it is hardest on people with hearing loss.
BBC TV audibility survey, 2011. RNID survey of members with hearing loss, 1998.
Adding music to a lesson made people remember less of it. Lectures and tutorials work better without it.
Moreno and Mayer, Journal of Educational Psychology, 2000.
Some people keep away from instrumental music for faith, a period of mourning or grief. nomus lets them keep the words.
Music makes what you see feel stronger, and sounds help short videos spread. Turn it off and judge what is being said.
Baumgartner et al., International Journal of Psychophysiology, 2006. TikTok, 2020.
Twitch reports that over 99% of its music takedowns were for background tracks. nomus removes most of that track. It does not grant rights to anyone's video.
Twitch, Music-related copyright claims, 2020.
Many autistic people find everyday sound too loud. Less music also means fewer songs stuck in your head at night.
Williams et al., Ear and Hearing, 2021. Scullin et al., Psychological Science, 2021.
A small neural network listens to 10 ms of sound at a time and decides, for every frequency, how much of it is a voice. Speech and singing pass. Instruments fade out. In Chrome it runs as WebAssembly; the macOS, Linux and Android apps run the same engine and weights natively.
Chrome hands nomus the audio of the tab you chose, and nothing else.
A real-time audio thread feeds the engine in 480-sample frames.
Rust, compiled to WebAssembly with no imports: it cannot reach the network or disk.
The voices, 20 ms later.
// One GRU layer, one 10 ms frame. PyTorch gate order: r, z, n.
matvec(&g.w_ih, &self.x, &g.b_ih, &mut self.gates_i);
matvec(&g.w_hh, hid, &g.b_hh, &mut self.gates_h);
for j in 0..h {
let r = sigmoid(self.gates_i[j] + self.gates_h[j]);
let z = sigmoid(self.gates_i[h + j] + self.gates_h[h + j]);
let n = (self.gates_i[2 * h + j]
+ r * self.gates_h[2 * h + j]).tanh();
self.h_new[j] = (1.0 - z) * n + z * hid[j];
}Filtering happens on your computer or phone. There is no nomus server for sound to go to.
No account, no analytics, no cookies. nomus does not count you, the sites you visit or what you play.
The extension and this site declare a connect-src policy, so the browser blocks requests to any other server. The Android app has no internet permission at all.
The one request nomus makes: the macOS and Linux apps check for updates when they start. The request names the system and processor type and nothing else, not even the app's version. You can turn it off in the app.
Check it yourself: open DevTools, pick the Network tab, and play something. Read the privacy page or the extension's source
Free, with no account. Pick where you listen.
The Chrome extension, for Chrome 116+ on Windows, Mac, Linux and ChromeOS. Press ⌘Shift Y (Ctrl on Windows) on a tab to filter it.
Add to ChromeEvery app: browsers, Spotify, video calls, games. macOS 14.2 or newer.
Coming soon
Every app, through PipeWire. A .deb for Ubuntu 24.04, Debian 12 and similar systems.
Coming soon
Apps that allow audio capture, such as YouTube and most podcast apps. Android 10 or newer, with a one-time setup.
Coming soon
Windows: the Chrome extension works there. A Windows app is not planned: Windows can copy what apps play but cannot silence the original, so filtering the whole system would need a signed kernel audio driver.