TuxTalk or Myna?

Who's writing this

I'm Ron, and I make TuxTalk. I don't write the code, though. Claude, Anthropic's AI, writes it, and I decide what it does, test it on real machines and use it for nearly everything I type.

I hold Claude to one rule on the code. An experienced developer should be able to open any file and find code they'd respect, not an AI mess, so it has to be easy to read and there can't be anything left lying around that isn't used. Every change also has to get through over 3,700 automated tests. There's more on that on the About page, under Keeping the code clean.

So keep in mind this comparison is written by the guy who makes the other app. I've tried hard to be fair to Myna, and every fact about it below links to where I got it, mostly Canonical's own announcement and Myna's code and release notes.

What Myna is

Myna is Canonical's speech-to-text app for Ubuntu Desktop. You press a shortcut, talk, and what you said goes into the app you're in. It runs on your own computer, and once a model is downloaded it doesn't need the internet and your audio isn't uploaded or stored (Ubuntu blog). It's free and open source under the AGPL.

Canonical calls it a technology preview and says to expect rough edges. You add it from a PPA, and its setup then pulls in the app, a speech model and a GNOME Shell extension as snaps from the Snap Store's edge channel (0.2.4 release notes).

Officially it's for Ubuntu 24.04, 26.04 and 26.10 on GNOME with Wayland (Ubuntu blog). Version 0.2.2 added Xubuntu, which its developer calls very experimental (issue #63). On KDE Plasma it stops with "Typing unavailable" for now, and they've said getting KDE working is a goal (issue #65).

Where Myna is better

You see the words while you're talking. Myna shows the text in the field as you speak (its settings). The closest TuxTalk gets is Type at pause in TuxTalk Labs, where each phrase is typed as soon as you pause while you keep on talking. Without it, TuxTalk types when you let go of the key.

Real word-by-word live text is something I'd look into if enough people want it. If that's you, say so in Feature Requests.

It's Canonical's. Canonical calls it an optional feature of Ubuntu 26.10 (Ubuntu blog), and they've put out five releases between October 7 and 19 (releases). If you're on stock Ubuntu with GNOME and want the one Ubuntu itself stands behind, that's Myna.

The speech models run in strictly confined snaps (README), and the Whisper one doesn't even get the microphone, the app hands it the audio (Whisper snap). And Myna starts with your desktop's own shortcut. TuxTalk isn't sandboxed like that, and to make hold-to-talk work on every desktop it installs a small hotkey helper that reads the keyboard, which asks for your password once during setup.

Myna's windows already come in a bunch of languages, 0.2.4 alone added ten (0.2.4 release notes), and its setup talks to screen readers (0.2.1 release notes). TuxTalk's windows are English only.

It also has a third speech engine, FunASR's SenseVoice, for Chinese, Cantonese, Japanese and Korean (FunASR snap). TuxTalk uses Whisper for those.

Where TuxTalk is different

With TuxTalk you hold F8, talk and let go. Myna had hold-to-talk and dropped it in 0.2.0, so now you tap its shortcut to start and tap again to stop (0.2.0 release notes). TuxTalk does that too, on F9, if you'd rather.

It isn't tied to one desktop or one distro. It works on KDE, GNOME, COSMIC, Sway, Hyprland, Cinnamon and other X11 desktops, and there's a .deb, an .rpm, an AppImage and an AUR package, so Mint, Fedora and Arch are covered as well as Ubuntu.

It isn't just for NVIDIA. Myna's graphics card engine is NVIDIA only, and for Parakeet it wants a Turing or newer card with more than 4 GB and driver 580 or later (Parakeet snap). Canonical says support for more is planned (Ubuntu blog). TuxTalk runs Parakeet and Whisper on AMD, Intel or NVIDIA cards through Vulkan, and on the processor if there's no card it can use.

You can teach it your words. A custom word list is the very first issue on Myna's GitHub, and its developer says it's their number one request (issue #1). TuxTalk has Your Words for names it should know, and Fix a Word for one it keeps getting wrong.

Canonical is keeping Myna to dictation on purpose. It isn't a voice assistant and doesn't control the desktop by voice (Ubuntu blog). TuxTalk takes "comma", "new paragraph" and "press enter", and in TuxTalk Labs you can say "edit that, make it more formal" right after you dictate. It can also turn a recording or a video into text with Transcribe a File.

Which one?

If you're on Ubuntu with GNOME and you'd like to watch the words show up as you talk, give Myna a try. It's early, but Canonical is clearly putting real work into it.

If you're on KDE or another desktop, or another distro, or you've got an AMD or Intel card, or you just want to hold a key and talk, that's what TuxTalk is for. And nothing stops you from trying both.

Checked on October 20, 2026, against Myna 0.2.4 and TuxTalk 0.17.0. Myna is moving fast, so some of this will go out of date. If you spot something that's wrong, tell me in Discussions and I'll fix it.

Also side by side: Dragon and Wispr Flow and Superwhisper.

Get TuxTalk   See what it does   Every Linux dictation app compared