// OMG! UBUNTU! — LINUX & OPEN SOURCE
Ubuntu’s ‘Myna’ AI lets you talk instead of type – but how does it work
Ubuntu is adding AI features this year, and founder Mark Shuttleworth hopes the distro will become the OS for the ‘agentic’ era. But big ambitions start with small seeds, and the first to be planted is a speech-to-text tool called Myna.
Age: Minus 4 months (it’ll debut in Ubuntu 26.10, out in October).
Appearance: A keyboard shortcut you press to avoid using your keyboard for typing.
What’s this about? A “lightweight speech-to-text application” powered by AI. You press a hotkey to activate it, talk at your computer and, like magic, your words appear on screen. Canonical’s VP of Engineering Jon Seager says any text field you can type in, you can talk into (or, in my case, talk at).
Yay, AI. What’s wrong with typing? Seager’s answer, delivered at the Ubuntu Summit: “Why type like an animal to your agent when you can just talk to it?”.
—Animal? I type gracefully, thank you! Good, cos you’ll your fingers to backspace through whatever the audio transcription model decides you said when you dictate an e-mail with your mouth was full of doughnut.
I have been told I talk like an animal… No comment.
AI, though… We’re not talking a conversational chatbot fused into the GNOME panel, nor a creepy “copilot” rifling through your root privileges. This is voice dictation powered by open speech recognition models. Decent dictation on Linux has been demanded for decades. Here, Canonical is set to actually deliver it.
Will Sam Altman be able to use my voice to train Cylons or something? No, your voice goes nowhere. Ubuntu will use a local, open model that runs on your device. No cloud AI service used. Your mic only wakes up when you press the relevant hotkey and audio is then processed in memory before being discarded. At least, that’s the plan.
—the plan? I’m hedging as the Myna GitHub is aims and diagrams right now (hence the setup using a flowery metaphor about seeds and planting). It walks through the aim: an Inference Snap, sandboxed, processes audio and then Myna, the speech orchestrator, manages the rest. Right now there’s no actual code to vet or test Myna out.