# Install the speech model when you need it

[Journal](/journal/)

August 2, 2026 · [guides](/journal/#guides)



Muniment Desktop discloses the speech model requirements before a user-triggered install and reports the resulting state without byte-level progress.

![A model package follows a green route into an installation bay beside five requirement and state markers.](/_astro/speech-model-install.YzrfwcY9_Z19Tkdn.avif)

A model download should follow an informed choice. Muniment Desktop puts that choice at the Voice action and discloses the installation requirements before transfer.

Our pinned Desktop evidence covers that boundary. It describes what the interface presents, what starts installation, and what the user can do after installation.

## Voice starts the choice

Muniment does not download the speech model in advance. A user presses Voice and selects Install to start the download.

The install card shows size, source, licenses, disk space, and install state. Those details keep the requirement and the current result together.

The interface shows no byte-level progress bar. Its install state reports the boundary without claiming granular transfer progress.

| Moment | What happens |
| --- | --- |
| Before installation | The install card shows size, source, licenses, disk space, and install state |
| Start | The user presses Voice and selects Install |
| After installation | The user presses Voice, and dictation starts |

## Installation ends at an available action

After installation, pressing Voice starts dictation. This is the only post-install behavior covered by the source.

The evidence does not evaluate inference, audio quality, or device performance. It also makes no claim about byte-level progress because the interface provides none.

The record comes from our Muniment Desktop installation note pinned to commit `180b5bbd030959319a250a85f27fd6b37aa55ac8`.

The interface asks before it installs, shows the requirements, and stops its evidence at the installed action.