1. Find out which session Ubuntu started
Open a terminal and run echo $XDG_SESSION_TYPE. It prints wayland or x11, and that one word decides which typing helper a dictation app can use on your machine.
On Ubuntu 22.04 and 24.04 the login screen still offers an “Ubuntu on Xorg” session. Canonical removed that option in Ubuntu 25.10, and in 26.04 LTS the Ubuntu Desktop session runs only on Wayland; X11 programs keep working through XWayland, and other desktops such as Xfce or MATE can still start on Xorg. Canonical made Wayland the default for NVIDIA graphics only in 24.10, so a 24.04 machine with NVIDIA drivers may be on Xorg. Check rather than assume.
- echo $XDG_SESSION_TYPE prints wayland or x11.
- 22.04 and 24.04: choose “Ubuntu on Xorg” at login if you want an X11 session.
- 25.10 and 26.04: the Ubuntu GNOME session is Wayland only.
2. Install the helper that fits your release
For an Xorg session, sudo apt install xdotool is enough. The package sits in universe on every current release, and its description is simply “simulate (generate) X11 keyboard/mouse input events”; xdotool’s own README warns that typing does not work on Wayland.
On GNOME Wayland, skip wtype even though Ubuntu packages it: it needs the Wayland virtual-keyboard protocol, which GNOME’s compositor, Mutter, does not implement. Handy’s README gives Ubuntu 26.04 users the same warning and points them to ydotool. ydotool and dotool work underneath the compositor through the kernel’s uinput device, so GNOME accepts their keystrokes.
Versions differ by release. Ubuntu 22.04 and 24.04 package ydotool 0.1.8 with the daemon in a separate ydotoold package; Ubuntu 26.04 packages ydotool 1.0.4, the rewrite in which the ydotoold daemon is mandatory, and upstream notes that ydotoold needs /dev/uinput, which usually means root. dotool is not in the Ubuntu archive as of October 2026: with gcc, go, libxkbcommon-dev and scdoc installed you build it from source, then join the input group so its udev rule grants access to /dev/uinput. That membership lets programs you run create keyboard input, so grant it only on a machine you control.
- Xorg session: sudo apt install xdotool
- 22.04 or 24.04 on Wayland: sudo apt install ydotool ydotoold
- 26.04 on Wayland: sudo apt install ydotool
- dotool from source: ./build.sh && sudo ./build.sh install, then sudo udevadm control --reload && sudo udevadm trigger
- dotool permission: sudo usermod -a -G input $USER, then reboot
3. Install IraVoice: the AppImage or the .deb
IraVoice 1.2.2 ships for Linux in beta as an AppImage that runs on Ubuntu 22.04 or later and updates itself. The quickest install is one command in a terminal, curl -fsSL https://iravoice.com/install.sh | bash, which puts it at ~/.local/bin/iravoice, adds it to the applications menu and opens it; or download the AppImage from iravoice.com/download, make it executable with chmod +x, and run it. If you would rather apt own it, download the .deb and, in the folder that holds it, run sudo apt install ./IraVoice-*.deb. Installing through apt rather than dpkg -i lets Ubuntu pull in the two libraries the package depends on, libgtk-layer-shell0 and libopenblas0, and puts the program at /usr/bin/iravoice; the .deb doesn’t update itself. There is no Snap or Flatpak build.
On first launch IraVoice downloads its Parakeet Unified EN speech model (about 700 MB) once; after that, recognition runs on the computer with NVIDIA Parakeet and keeps working offline. The 3-day free trial begins with your first dictation and needs no email, card or account. IraVoice Local then costs $79 once for one computer, $109 for two, or $149 for three.
IraVoice chooses a typing helper by itself: xdotool on an Xorg session, and dotool or ydotool on GNOME Wayland, which need the /dev/uinput access from the previous step. Wayland support is limited in the beta, so if text does not appear, test the helper on its own before changing anything in IraVoice.
4. Bind iravoice --toggle-transcription to a GNOME shortcut
IraVoice’s own key on Linux is Ctrl+Space: hold it to talk, or tap it for hands-free dictation and tap again to stop. On Wayland that global shortcut may not reach the app, so give GNOME a custom shortcut that runs iravoice --toggle-transcription, which starts or stops a dictation in the running app. If the shortcut can’t find iravoice, use its full path, for example /home/you/.local/bin/iravoice --toggle-transcription.
GNOME’s help describes the same steps for any command and notes that the command must be valid, so run it once in a terminal while IraVoice is open. Because the shortcut toggles, press it once to start speaking and again to have the text typed.
- Open Settings → Keyboard and select View and Customize Shortcuts.
- Select Custom Shortcuts, then Add Shortcut (or + if one already exists).
- Name it IraVoice dictation and enter the command iravoice --toggle-transcription (or the AppImage’s full path followed by --toggle-transcription).
- Click Add Shortcut…, hold down the key combination, then click Add.
5. Choose the microphone in Settings → Sound
Recognition can only be as good as the input. Open Settings → Sound and pick the device in the Input section; GNOME’s help says the level indicator should respond as you speak, and that USB microphones appear as separate devices you may need to select by hand. Ubuntu 26.04’s Sound settings separate input and output more clearly than earlier releases.
Before dictating into anything important, say one short sentence into Text Editor. If the level meter moves but no text appears, the problem is the typing helper; if the meter stays flat, it is the device or its mute switch.
6. What Myna changes in Ubuntu 26.10
Canonical announced Project Myna on June 17, 2026 as offline dictation for Ubuntu Desktop: press a shortcut, speak, and the text appears in the application you are using. The first release targets Ubuntu 26.10 on Wayland with GNOME as the validated desktop, and leaves voice commands, desktop control, translation and automatic language detection out of scope. Ubuntu 26.10 is in beta and scheduled for October 15, 2026.
Myna is not a single package. Its AGPL-licensed repository describes a client snap plus one inference snap per model family, and a Myna Settings app, shipped as a deb, whose setup wizard installs the client and a model from the Snap Store. OMG! Ubuntu’s August hands-on found that an early build started and stopped with Super+T and handed text to apps through IBus, and reported that the feature is expected to be opt-in.
Canonical’s announcement is about 26.10 and does not promise Myna for 22.04, 24.04 or 26.04 LTS. If you stay on an LTS release, an app plus the right typing helper remains the way to dictate.