What Ubuntu's Myna preview keeps on the machine

What happened

On October 12, 2026, Jean-Baptiste Lallement wrote on the Ubuntu blog that Myna, a speech-to-text technology preview, is available. Canonical says it is optional, a preview, and not a finished product. Speech recognition runs on the machine. Once a model is downloaded, transcription does not require the internet, and audio is not uploaded or stored. The engines named are Parakeet, Whisper, and FunASR. The blog says the preview is for Ubuntu 24.04 LTS, 26.04 LTS, and 26.10, on amd64 and amd64v3, with GNOME on Wayland. NVIDIA GPU acceleration is offered if a GPU is present. Other accelerators are planned. Dictation is the feature. Canonical says Myna does not control the desktop by voice. The paper filing the preview isFutureweb.

Where the install path already disagrees

The Ubuntu blog describes a package path. Phoronix's October 12 account says the preview can be installed on the LTS releases via Snap, and it quotes Canonical that the core experience is a shortcut, speech, and text in the application you are using. OMG Ubuntu's same-day account also describes a local model and GNOME on Wayland, and it says other desktops are planned. This page does not print install commands, and it does not pick the blog's package path over Phoronix's Snap sentence. Both are in the record until Canonical says one of them is wrong. The desk isOpen Source.

What the preview refuses to be

A voice assistant that drives the desktop is the thing Canonical says this preview is not. A finished product is the other thing it says Myna is not. OMG Ubuntu adds that the project is under active development and that features may change. Local recognition is the claim that survives those limits. What a license obligates, for the code Canonical points at, iswhat a license obligates.

Why it matters on this desk

The dependency is where the audio goes. Canonical's sentence is that it stays on the machine after the model download, and that it is not uploaded or stored. That is a different fact from a cloud dictation service. ONLYOFFICE's split of AI out of an open-source build is a different open-source document from the same week,what ONLYOFFICE Docs 10.0 changes in the open-source build.

What this page will not import

The blog includes install commands. This page does not repeat them. It also does not treat a technology preview as the default desktop. A session that cannot outlive the desk it was made for is a privacy fact beside this one,what a session cannot be.

Where the accounts overlap

The Ubuntu blog, OMG Ubuntu, and Phoronix overlap on the October 12 preview, the local model, the three engine names, and support for Ubuntu 24.04 LTS, 26.04 LTS, and 26.10 with GNOME. They do not overlap on a single install mechanism. The blog and Phoronix describe different delivery paths, and this page leaves that split in place. Other accelerators are planned. They are not in the preview Canonical is offering now. The source repository Canonical names is the public myna project. This page does not treat a preview tag as a stable release. The source repository Canonical names is the public myna project. This page does not treat a preview tag as a stable release. The source repository Canonical names is the public myna project. This page does not treat a preview tag as a stable release.

What to watch next

A Canonical note that picks one install path, adds a desktop besides GNOME on Wayland, or withdraws the not-uploaded claim would change the record. Another hands-on that repeats the preview would not. What on-device inference keeps, when the model is the point, iswhat on-device inference keeps.

What would change the record

A statement that audio is uploaded, or that Myna now controls the desktop, would change it. A stable release that replaces the technology-preview label would change it. A command list would not. Jean-Baptiste Lallement is the author on the Ubuntu blog. The source repository Canonical names is the public myna project. This page does not treat a preview tag as a stable release. OMG Ubuntu says a local model does not send audio off the machine. That matches the blog. The install disagreement does not. OMG Ubuntu says a local model does not send audio off the machine. That matches the blog. The install disagreement does not. OMG Ubuntu says a local model does not send audio off the machine. That matches the blog. The install disagreement does not.

What stays on the machine

What stays specific is an optional October 12, 2026 technology preview: recognition on the computer, no internet required after the model download, audio not uploaded or stored, engines named Parakeet, Whisper, and FunASR, Ubuntu 24.04 LTS, 26.04 LTS, and 26.10, GNOME on Wayland, dictation only. NVIDIA acceleration is offered when a GPU is present. The install path is not settled between the accounts this page is using. Jean-Baptiste Lallement is the author on the Ubuntu blog. Other accelerators are planned. They are not in the preview Canonical is offering now. Jean-Baptiste Lallement is the author on the Ubuntu blog. Other accelerators are planned. They are not in the preview Canonical is offering now. Jean-Baptiste Lallement is the author on the Ubuntu blog. Other accelerators are planned. They are not in the preview Canonical is offering now.