kyrr
GitHub

Noise suppression and dictation for Windows and Linux

kyrr is a free, open-source app. It removes background noise from your microphone and converts speech to text in German, English and more European languages. All audio is processed on your computer.

Download README on GitHub ›
kyrr's window: noise filter, improve voice and dictation, each with one switch and a gear for its settings

Hold F9, speak, release: the text appears where the cursor is.

Features

Noise suppression

Removes background noise such as typing, fans or traffic from the microphone signal in real time. kyrr uses the neural network DPDFNet, with a large model and a small one that needs less processing power. Apps such as Teams, Zoom, Discord, OBS or a browser receive the cleaned sound without any change to their settings.

Improve voice

Filter for low rumble, tone presets (Neutral, Warm, Clear, Bright or custom), compressor, de-esser, pop protection and automatic volume. Default: neutral tone, medium compression, light de-essing, rumble filter and automatic volume. Everything can be changed under the voice gear.

Dictation

Hold F9 while you speak. When you release it, kyrr types the text at the cursor, in the app you are typing in. German, English and more European languages, such as French, Spanish, Italian or Dutch. “New line” inserts a line break. While you dictate, a small wave at the bottom of the screen shows that kyrr is listening, with the words heard so far, and then that it is writing. Instead of F9 you can set a key of your own, or tap once to start and once to stop.

Voice memos

Hold F8, speak, let go. kyrr saves the recording in the folder Documents/kyrr and names it after its first words. In the list you can play, rename or delete a recording. One click turns a recording into text, with the time at which each part starts. Voice memos can be switched off on the main page.

Level check

Meters show what the microphone picks up and what apps receive. A ten-second test records your voice and plays it back before and after processing. You can also listen to yourself live on headphones.

How it works

  1. Installation. Windows: an installer, which needs administrator rights. Linux: a single AppImage file, no administrator rights. Omarchy: a plugin for the bar, installed with one command. kyrr sets itself up on first start.
  2. Audio. On Windows, kyrr adds a filter to the Windows audio system that runs directly on the microphone. On Linux, kyrr creates a PipeWire microphone called “kyrr” and makes it the default. Your apps need no changes. If you switch to another microphone, kyrr switches with it.
  3. Dictation. The first time you dictate, kyrr downloads the speech model for your language once (German 900 MB, English 663 MB, other European languages 900 MB). After that, dictation works offline.
  4. Updates. kyrr checks once a day for a new version and tells you about it. On Windows, “Update” installs it. On Linux, run the install commands again; a running kyrr is replaced by the new version.

Privacy

Audio and text are processed only on your computer. kyrr needs no account and sends no usage data.

kyrr connects to the internet for three things: downloading the speech model once (each file checked against a SHA256 hash), a daily update check with GitHub (can be switched off in Setup), and downloading an update when you click “Update”. Voice memos and their text stay in your Documents folder. Details in the README

System requirements

Windows

  • Windows 10 (22H2) or Windows 11, 64-bit
  • Intel or AMD processor. On Windows on ARM, only dictation works.
  • Administrator rights for installing

Linux

  • 64-bit Intel or AMD, glibc 2.39 or newer: Ubuntu 24.04, Fedora 40, Debian 13, Linux Mint 22, Arch, or newer
  • PipeWire. If it is missing, kyrr offers to install it.
  • Noise suppression needs a processor with AVX2 (most from 2015 onwards).

Both: no graphics card needed. Disk space for the speech model: 663 MB (English) or 900 MB (German and other European languages).

Download

Free of charge. Three versions: Windows, Linux and Omarchy. Source code on GitHub under MIT or Apache 2.0.

Windows

Download for Windows

The installer is not code-signed yet, so Windows SmartScreen shows a warning: click “More info”, then “Run anyway”. Why it is not signed, and how to verify the file

Start at sign-in

kyrr starts automatically when you sign in, once its first-start setup is done. To change this: Setup in kyrr, “Start with Windows”.

Linux (beta)

The Linux version is still a beta: it works, but has been tested on fewer systems than the Windows version. On Omarchy, use the Omarchy tab instead.

Paste this into a terminal. It downloads kyrr to ~/.local/bin, makes it executable and starts it. Run it again for an update: a running kyrr closes and the new one starts.

mkdir -p ~/.local/bin
curl -fL -o ~/.local/bin/kyrr.new https://github.com/heast13/kyrr/releases/latest/download/kyrr-x86_64.AppImage && chmod +x ~/.local/bin/kyrr.new && mv -f ~/.local/bin/kyrr.new ~/.local/bin/kyrr
~/.local/bin/kyrr --replace &

Without a terminal: download the AppImage, allow it to run as a program in your file manager (right click, Properties) and double click it. On first start kyrr adds itself to your app menu. On GNOME 48 or newer and KDE the desktop asks once for the dictation key and once whether kyrr may type; kyrr explains both steps. Notes for other desktops

Start at sign-in

kyrr starts automatically when you sign in, once its first-start setup is done. To change this: Setup in kyrr, “Start when you sign in”.

Uninstall

Quit kyrr (“Quit”, bottom left), then paste:

rm -f ~/.local/bin/kyrr \
  ~/.config/autostart/io.github.heast13.kyrr.desktop \
  ~/.local/share/applications/io.github.heast13.kyrr.desktop \
  ~/.local/share/icons/hicolor/256x256/apps/io.github.heast13.kyrr.png \
  ~/.local/share/icons/hicolor/scalable/apps/io.github.heast13.kyrr.svg
rm -rf ~/.config/kyrr ~/.local/share/kyrr

The last line also deletes your settings and the speech models. Leave it out to keep them.

Omarchy

For Omarchy 4.0 or newer. kyrr is a plugin for the bar: the same app, with its menu in Omarchy's style and no window of its own. Paste this into a terminal (the same line updates the plugin later):

omarchy plugin update io.github.heast13.kyrr --yes || omarchy plugin add https://github.com/heast13/omarchy-kyrr.git --enable

Then click the kyrr icon in the bar. The menu installs kyrr for you and starts its setup. Details on GitHub

Made with Slint

Questions

Does kyrr need an internet connection?

Only to download the speech model the first time you dictate, for the daily update check, which you can switch off, and to download an update when you click “Update”. Noise suppression works without internet.

How long can a dictation be?

Up to 15 minutes. Longer recordings are cut off at that point.

Why does dictation not work in some windows on Windows?

Windows does not let normal programs see the keyboard while a window running as administrator is in front, such as a terminal opened with “Run as administrator”. kyrr does not react there. Click into another window and dictate there.

Which apps get the cleaned sound?

On Windows, every app that uses the microphone. On Linux, every app that uses the default microphone, which kyrr sets to its own “kyrr” microphone.

Is there a Mac version?

No. kyrr is available for Windows and Linux.

How does kyrr compare with NVIDIA Broadcast and Krisp?

NVIDIA Broadcast needs an NVIDIA RTX graphics card. Krisp needs an account and is free for 60 minutes a day, with a subscription beyond that. Both work as a virtual microphone that you select in each app. kyrr needs no selection in the apps, because it filters the microphone itself on Windows and sets its own microphone as the default on Linux. It needs no account and includes dictation. As of October 2026; corrections welcome.

Is kyrr finished?

kyrr is in active development. Found a problem or have an idea? In kyrr, Setup, About, “Feedback” opens an e-mail to stueckig.habicht_2o@icloud.com. Bug reports and suggestions are also welcome on GitHub.