Ubuntu GNOME and Fedora KDE
Experimental setup for Ubuntu 26.04 GNOME and Fedora 44 KDE Plasma. Not yet verified on real hardware.
Warning
Experimental, and not yet verified on real hardware. The GNOME and KDE support is built and tested in containers and headless sessions only. Nobody has run it on a physical Ubuntu 26.04 or Fedora 44 desktop with a real microphone, so expect rough edges. The packages are built but not published, so the apt and rpm commands below won't work until they are.
This page covers the Wayland sessions on Ubuntu 26.04 (GNOME) and Fedora 44 (KDE Plasma). X11 sessions aren't supported. On Omarchy and Hyprland, follow the Linux guide instead.
On these desktops Mouthtype doesn't use a compositor config. It asks the desktop portal for two things, typing access and global shortcuts, and you approve each once.
Install
Two packages. mouthtype-linux is the daemon and command line. mouthtype-linux-desktop adds the settings window, tray icon and, on KDE, the floating recording pill. Install both on GNOME and KDE.
Ubuntu 26.04 (deb):
sudo apt install mouthtype-linux mouthtype-linux-desktop
mouthtype-linux-setupThis needs the Mouthtype apt source from the Linux guide, which isn't live yet. Until then, build the packages from packaging/debian.
Fedora 44 (rpm):
sudo dnf install ./mouthtype-linux-*.x86_64.rpm ./mouthtype-linux-desktop-*.noarch.rpm
mouthtype-linux-setupThere's no COPR repo yet, so build the rpms with packaging/rpm/build-rpm.sh. The core package needs ffmpeg (Fedora's own or RPM Fusion's), pw-record and notify-send.
Run mouthtype-linux-setup as your normal user, never with sudo. It downloads the speech model (about 465 MiB), runs the doctor check, enables the mouthtype-linux user service and adds the settings window to your login autostart so the tray icon is there after you sign in.
First run and the two prompts
Setup shows two prompts. Both come from your desktop, not from Mouthtype.
- Typing permission (Remote Desktop). During setup the desktop asks whether Mouthtype may control the keyboard. This is how text reaches the app you're in. Approve it and the grant is stored, so you aren't asked again. If you decline, setup carries on and dictation still records. Your text is saved for recovery instead of typed. Ask again with
mouthtype-linux-setup --grant. - Global Shortcuts. When the service starts, the desktop shows its shortcut dialog with Mouthtype's four actions. Accept the suggested keys or pick your own.
The desktop reports back what it actually granted, which can differ from the suggestion if a key was taken. Check the shortcut settings in GNOME Settings or KDE System Settings if a key doesn't work.
Shortcuts
| Action | Suggested keys |
|---|---|
| Dictate while held | Ctrl+Alt+M |
| Start or stop dictation (toggle) | Ctrl+Alt+Space |
| Cancel the current take | Ctrl+Alt+Escape |
| Transform the selected text | Ctrl+Alt+R |
Hold Ctrl+Alt+M, speak, and let go. Or tap Ctrl+Alt+Space to start and again to stop. If you let go of a held key and the desktop never reports the release, the take ends anyway.
If the Global Shortcuts portal is missing or you refuse it, GNOME falls back to a single custom keyboard shortcut that runs mouthtype-linux toggle. A command shortcut has no key release, so the fallback is toggle only.
Where the text goes
Delivery starts recovery-only: a take is saved and nothing is typed. To have Mouthtype paste or type into the app you're using, switch the delivery mode in the settings window or set delivery.mode to paste or type in linux.json. The same opt-ins as the other Linux setups apply: either list exact allowed_classes or set any_window, and set allow_unknown_secure_field to true. See the setup guide for the keys.
Two behaviors worth knowing:
- Password fields. Mouthtype asks the accessibility bus (AT-SPI) what has focus. If the focused widget is a password field, delivery is refused for typing and pasting alike, and the text goes to recovery. This is best effort. An app that doesn't report focus over AT-SPI shows up as unknown, and the
allow_unknown_secure_fieldsetting decides what happens then. - Clipboard restore. Paste puts the transcript on the clipboard, sends Ctrl+V, then puts your earlier text clipboard back about 1.5 seconds later. It skips the restore if you copied something else in between, and it never restores an image, a file list or a password manager's copy. Turn it off with
delivery.restore_clipboard.
Tray icon and pill
The tray icon shows state and has Start or Stop dictation, Open Mouthtype, Settings and Quit.
- GNOME has no tray of its own. The icon needs the AppIndicator extension. Ubuntu ships it and enables it by default, and
mouthtype-linux-desktoprecommendsgnome-shell-extension-appindicator. On Ubuntu 26.04 the extension isubuntu-appindicators@ubuntu.comfromgnome-shell-ubuntu-extensions, whichubuntu-sessiondepends on and which provides thegnome-shell-extension-appindicatorname, so the recommendation resolves there (checked against the 26.04 archive, 2026-11-02). If the icon is missing, check the extension is on. The tray also appears if the extension starts after Mouthtype does. - KDE Plasma shows the icon natively, and also the floating recording pill. The pill needs the layer-shell module (
qml6-module-org-kde-layershellon Ubuntu,layer-shell-qton Fedora). Without it you get the tray and settings window only. GNOME has no pill.
Open the settings window from the tray, or run mouthtype-desktop. mouthtype-desktop --background starts tray only, which is what the autostart entry does.
Check your setup
mouthtype-linux doctorIt's read-only. It reports the detected desktop, which delivery and shortcut backends were chosen, whether the portal answered, the microphone, the speech files, and whether the service is running. Exit status 0 means the required checks passed, 2 means something needs attention. Add --json for structured output. Passing doesn't prove the microphone works or that text reaches a particular field.
| Message | What it means and the fix |
|---|---|
App info not found | The portal couldn't find a desktop file for Mouthtype. Make sure the package installed app.mouthtype.Mouthtype.desktop and that its Exec program is on the portal's PATH, then run update-desktop-database and sign out and back in. |
| Portal could not be used | xdg-desktop-portal or your desktop's backend isn't running. Install xdg-desktop-portal-gnome or xdg-desktop-portal-kde, and check systemctl --user status xdg-desktop-portal. |
| Portal client could not be imported | Install python3-jeepney, which has the same name on Ubuntu and Fedora. |
| No tray icon on GNOME | Install and enable the AppIndicator extension, then sign out and back in. Check with gnome-extensions list --enabled. |
| Typing never happens, text only lands in recovery | The typing grant is missing or was revoked. Run mouthtype-linux-setup --grant, or read the take with mouthtype-linux list and recover ID. |
| Desktop reported as unknown | You're on an X11 session or an unsupported compositor. Log in to a Wayland session. |
| Service is not answering | Start it with systemctl --user start mouthtype-linux, or run mouthtype-linux serve in a terminal to see why it stops. |
Uninstall
Run this before removing the package:
mouthtype-linux-setup --uninstallIt removes the user service, the autostart entry and the GNOME fallback shortcut. Models and recordings stay. Add --purge to delete them and the stored typing grant too. To forget the grant on the desktop side, remove Mouthtype in GNOME Settings or KDE System Settings. Then remove the packages:
sudo apt remove mouthtype-linux-desktop mouthtype-linux # Ubuntu
sudo dnf remove mouthtype-linux-desktop mouthtype-linux # Fedora