Want to try Ubuntu’s AI dictation feature? Here’s how

Ubuntu’s new AI-powered dictation tool Myna isn’t officially ready, but you can give it an early test.

Canonical has added the Myna orchestrator snap and a pair of speech-recognition AI models to the Snap Store recently, and made its new desktop Myna Settings tool available to test from a PPA.

Ubuntu 26.10 also comes with another component preinstalled, as the gnome-shell-ubuntu-extensions package has a GNOME Shell extension that triggers an on-screen HUD when the Myna local AI speech-to-text tech is listening or transcribing.

While a formal call for testing hasn’t been made, all the pieces needed to test Myna are available – for those who care to put them together.

If that includes you, read on for how to do it – but don’t expect things to work perfectly just yet, mind.

How to try Myna on Ubuntu

You can install Myna on Ubuntu 26.04 LTS or 26.10, but the Myna Shell (which triggers the on-screen HUD) is only packaged for GNOME 51 and 26.10 (notifications get used to convey activity instead).

The on-screen HUD only shows when the trigger key activates Myna and it is listening or transcribing. It’s been substantially reworked from the version I previewed back in August. It now triggers a myna.hud process rather than drawing it on screen itself.

4 different visual styles are available for the HUD: default (a plain bar using the system accent colour); vumeter (think volume), ribbon (but as animated as the original one, alas) and bar (same to default but, in my test, lacks a trough).

To test it, you need the following things

• myna-config desktop app from a PPA
• myna snap from the Snap Store
• myna-* AI model from the Snap Store

If you’re on Ubuntu 26.10:

• Myna Shell extension (it’s preinstalled)

Her’s how to install each in turn.

Install the Myna Settings app by adding the following PPA (maintained by a Canonical engineer working on Myna) which provides packages for Ubuntu 26.04 LTS and 26.10:

sudo add-apt-repository ppa:charles05/myna-config

sudo apt install myna-config

Enable an experimental snap feature Myna requires before it will install:

sudo snap set system experimental.user-daemons=true

Install the Myna speech-to-text client from the Snap store:

sudo snap install myna --edge

Finally, you need to install “brains” of the operation: an AI model to transcribe audio to text.

If you open the Myna Settings (myna-confg) setup tool now it will suggest a model suitable for your hardware. But if you click on the ‘install’ button it will throw an error (it points to stable versions), so you’ll need to install the model by hand.

Whisper and Parakeet are two local AI models already on the Snap Store, but Funasr is expected to join them soon. The models run entirely locally, on your device, but aren’t small. You’ll need 1-2GB of free disk space¹, depending on the one you choose.

Install Parakeet, NVIDIA’s fast, open-weight ASR model supporting 25 languages and punctuation:

sudo snap install myna-parakeet --edge

Or install Whisper, OpenAI’s multilingual, open-source model:

sudo snap install myna-whisper

Now open Myna Settings and it should skip past the setup steps and show you, well, settings.

I suggest you test Myna before you poke around. Set a hotkey shortcut in the config app², then focus a text field (I tend to use Text Editor) and press it.

When activated, the HUD (26.10) or a notification (26.04) will appear. Begin to talk and, if Myna says “listening”, keep rambling until you’re done, then press the same hotkey again. The model will then transcribe your spoken words to text, and those will appear in the focused app.

I have tried both Parakeet and Whisper. The former was better at accuracy (even spelling my surname correctly, which most don’t), while Whisper was faster. Options in the myna-config app does let change models (smaller or larger variants).

Is Myna any good?

Myna is a certainly a boost to Ubuntu’s accessibility, but it’s too soon to say if it’ll prove popular as a primary input method. Speech to text tools aren’t new, they’re just more accurate than they used to be. Myna requires a key press, so it only “listens” when you need it to.

It can’t be triggered with a “hey myna” – not yet, anyway.

Hot-key triggered dictation is intentional, so better suited to longer-form text entry than barking “firefox” at the App Picker. One annoying thing is that Myna loses focus easily (another app opens, you press a button or mouse away), losing your “spiel” entirely.

And there’ll be time spent proofing written speech, battling “too noisy” errors and other quirks – there is a reason most of us still type messages, even on phones that have long had ML-powered dictation tools built in.

Still, it’s better than what was previously available. It may not be as ‘clever’ as the voice typing features Google offers on its new Googlebook laptops, but it is local, offline and entirely in your control.

1. Snap is configured to keep one “good” backup of everything you install, so when it gets updated the only one stays behind, meaning each snap uses twice the amount of space. ↩︎
2. On one of my laptops Myna defaults to super + j but on another it uses super + t. You can see/change it in the Myna Settings app. ↩︎