You can try Ubuntu’s new Myna AI dictation tool – here’s how

For users running the development or stable releases of Ubuntu 26.10, the system also comes with a crucial preinstalled component through the gnome-shell-ubuntu-extensions package. This package delivers a specialized GNOME Shell extension designed to trigger an intuitive, on-screen Heads-Up Display whenever the local AI speech-to-text technology is actively listening or processing a transcription. While Canonical has yet to issue an official, formal call for testing, all the foundational pieces required to run Myna are out in the wild for those willing to assemble them. The feature has undergone significant evolution since its initial preview phase, transitioning into a robust offline solution.

Testing Myna requires specific system configurations depending on the exact Ubuntu release being utilized. On Ubuntu 26.10, the integration is more seamless due to the inclusion of the Myna Shell extension, which is currently packaged exclusively for GNOME 51 and the 26.10 environment. For users on Ubuntu 26.04 LTS, desktop notifications are leveraged instead to convey real-time system status and operational feedback. The on-screen HUD appears dynamically only after the user presses the designated trigger key to activate Myna, remaining visible during active listening and transcription phases, or when relaying potential error states such as high background noise levels or a sudden loss of window focus.

The visual feedback provided by the HUD offers multiple stylistic variations to suit user preferences. These include a default style featuring a clean bar utilizing the system accent color, a volume-meter style, an animated ribbon style, and a bar variant closely mirroring the default setup. To manage these visual preferences and overall system behavior, users can install the dedicated Myna Settings application. This utility is made available via an upstream PPA maintained directly by a Canonical engineer involved in the Myna development cycle, supplying packages tailored for both Ubuntu 26.04 LTS and Ubuntu 26.10.

Before installing the core Myna speech-to-text client from the Snap Store, users must enable a specific experimental snap feature that the software relies on to function correctly. Once this system-level configuration is adjusted, the Myna client can be fetched from the snap repository. However, to complete the installation and provide the underlying intelligence required for transcribing audio into text, users must also acquire a local AI model snap.

The Myna Settings application features a built-in first-run wizard intended to handle these installations automatically and recommend a model suited to the host hardware. Because the software remains in an active testing phase, utilizing this automated installation button can currently trigger an error. Consequently, early testers must install their preferred speech-to-text model manually via the command line.

The Snap Store currently hosts three primary AI models compatible with the architecture: Whisper, Parakeet, and Funasr. These models operate entirely offline and locally on the user’s device, ensuring complete privacy, though they require significant storage space ranging from one to two gigabytes depending on the selected option. Early evaluations indicate distinct trade-offs between the choices. Parakeet, an open-weight automated speech recognition model developed by NVIDIA that supports twenty-five languages and built-in punctuation, offers exceptional accuracy, frequently spelling complex surnames correctly where other models fail. Whisper, OpenAI’s multilingual open-source model, tends to operate with greater processing speed, though it occasionally introduces homophone transcription errors. Funasr represents another viable alternative available for deployment.

Once the preferred model snap is installed, the Myna Settings utility successfully bypasses its initial setup hurdle and displays the full suite of configuration options. Through this interface, users can adjust model sizes—modulating the parameter counts to favor either rapid processing on lower-end hardware or maximum accuracy to minimize errors and handle complex vocabulary.

With the components successfully deployed, verifying the system requires ensuring that the local microphone is fully operational within the system sound settings. Users must then assign a custom hotkey shortcut within the configuration app to serve as the physical trigger for the dictation feature. Activating the tool involves focusing on any text field within a desktop application—such as a standard text editor—and pressing the assigned shortcut.

Upon activation, the Myna HUD manifests on the desktop or via a notification, signaling that the system is ready. Speaking clearly allows Myna to process the audio stream, culminating in transcribed text being automatically populated into the active text field once the hotkey is pressed a second time to end the session. The software is sensitive to context; if focus shifts away from the target text field due to another application opening or a manual mouse click, the system registers a focus loss and interrupts the transcription process.

Myna represents a major step forward for Ubuntu’s accessibility infrastructure, offering robust, offline, AI-powered dictation capabilities that far surpass previous native offerings. Whether voice dictation will achieve widespread adoption in daily desktop workflows remains an open question, given that offline speech recognition has existed on mobile devices for years while many users still prefer traditional typing for day-to-day messaging and productivity tasks. Nonetheless, dictation excels at longer-form text entry, and because Myna operates entirely locally without sending audio recordings to external servers, it provides a secure and private alternative. As Canonical continues to experiment with desktop-agnostic technologies built on modern Wayland and Mutter stacks, Myna introduces a fresh dimension of intelligent interaction to the Ubuntu ecosystem.