Ubuntu Users Can Now Test Canonical’s New Local AI-Powered Dictation Tool, Myna

Canonical’s upcoming artificial intelligence-driven dictation system, Myna, is not yet officially ready for general release, but early adopters and curious Linux enthusiasts can already put the software through its paces. The development of Myna marks a significant step forward for speech-recognition technology within the Ubuntu ecosystem, bringing fully offline, local AI transcription capabilities to the popular desktop operating system.

Over the past several weeks, Canonical has steadily populated the Snap Store with the necessary backend infrastructure for the project. This includes the core Myna orchestrator snap alongside a selection of speech-recognition AI models. Additionally, a new desktop configuration utility known as Myna Settings has been made available for testing via a dedicated Personal Package Archive managed by a Canonical engineer.

Integration runs deep within the desktop environment for users running cutting-edge releases. Ubuntu 26.10 ships with a preinstalled component housed within the standard GNOME shell extensions package. This extension is designed to trigger a visual heads-up display on the screen whenever the local AI speech-to-text technology is actively listening for voice inputs, processing transcriptions, or reporting system states.

Although Canonical has not yet issued a formal, widespread call for community testing, all the individual pieces required to assemble and operate Myna are now publicly accessible for those willing to experiment. The feature has undergone noticeable evolution and refinement since it was first previewed to the public earlier in the year. While early testers can now experience the technology firsthand, developers advise that the software remains experimental and should not yet be expected to function flawlessly in daily production environments.

The architecture of Myna relies heavily on locally executed models, ensuring that user privacy is respected by design. Unlike cloud-based transcription services that route audio data to external servers, Myna processes all spoken words entirely on the host device. Audio clips never leave the user’s computer, nor are they permanently stored on local drives.

Depending on the specific operating system release being used, user feedback and system statuses are delivered differently. On Ubuntu 26.10, the on-screen heads-up display provides immediate visual feedback. Meanwhile, users running Ubuntu 26.04 LTS rely on standard desktop notifications to convey operational status, as the specialized shell extension is packaged explicitly for newer GNOME environments.

The heads-up display appears dynamically whenever the user presses the designated trigger key to activate the dictation service. It remains visible during listening and transcribing phases, and also serves to communicate error states, such as warnings regarding high background noise or the loss of active window focus. Several distinct visual styles are offered for the HUD interface, ranging from a default plain bar utilizing the system’s accent color to volume-style meters and ribbon animations, giving users a degree of visual customization.

Installing and configuring Myna requires setting up a few preliminary components. Users must first add the experimental PPA to install the Myna configuration utility. Furthermore, an experimental Snap feature governing user daemons must be enabled on the system before the primary Myna speech-to-text client can be pulled from the Snap Store in its edge channel.

The most critical components of the system are the underlying AI models that serve as the computational brains for speech translation. The Myna ecosystem supports prominent open-weight models such as OpenAI’s Whisper and NVIDIA’s Parakeet, with additional models like Funasr expected to join the lineup in the near future. These models are comprehensive and computationally intensive, requiring between one and two gigabytes of free disk space depending on the specific variant chosen.

Parakeet is recognized as a fast automatic speech recognition model supporting twenty-five languages alongside integrated punctuation, while Whisper offers robust multilingual, open-source transcription capabilities. Once the chosen model package is installed alongside the main client, the Myna Settings application recognizes the backend and presents users with a suite of configuration options.

Operating the tool involves assigning a preferred keyboard hotkey shortcut within the configuration utility. After focusing any standard text input field, such as a basic text editor, the user presses the shortcut to summon the transcription interface. Once the visual indicator confirms that the system is actively listening, the user dictates their message. Pressing the hotkey a second time signals the end of the speech segment, prompting the local AI model to translate the audio into text and automatically insert it into the active application window.

Initial evaluations of the software reveal notable differences in performance between the available models. Parakeet has demonstrated high levels of transcription accuracy, occasionally succeeding at complex tasks like spelling uncommon surnames correctly, whereas Whisper has earned praise for processing speed. Users retain the flexibility to swap between different models and adjust various parameters through the configuration app.

The introduction of Myna represents a substantial boon for accessibility on the Ubuntu platform, offering an offline, AI-powered dictation alternative that surpasses previous local speech recognition capabilities. However, broader questions remain regarding how widely everyday users will adopt dictation for routine tasks. While offline speech-to-text has been a staple of mobile operating systems for years, the vast majority of computer users continue to rely on traditional keyboards for composing messages, electronic mail, and text prompts.

Voice dictation inherently suits longer-form text entry far better than brief commands or application launching. The technology also demands careful attention to window focus, as accidental mouse clicks or opening secondary windows can disrupt the active transcription process and result in lost text. Because Myna operates strictly on-demand through manual activation and lacks any always-listening wake-word functionality, it prioritizes user privacy over ambient convenience.

As Canonical continues to refine the technology, Myna stands out as an ambitious experiment in bringing advanced, privacy-focused machine learning tools directly to the Linux desktop. While current development efforts are tightly integrated with the GNOME and Wayland display stack, the foundational technology is being built to remain desktop-agnostic, pointing toward a versatile future for speech recognition across the broader open-source ecosystem.