Open the preferences from the Extensions app, or with:
gnome-extensions prefs murmur@roman-16.github.ioEvery setting takes effect on the next dictation. Nothing needs a restart.
Your key for the Voxtral realtime endpoint. Murmur does nothing without it and says so when you press the shortcut.
It is stored like every other GNOME setting, in dconf, unencrypted. Privacy covers what that means.
How much audio Voxtral buffers before it transcribes. More context means better accuracy; less means text appears sooner.
| Preset | Value | Feels like |
|---|---|---|
| Instant | 240 ms | Words appear almost as you say them, with more corrections |
| Fast | 500 ms | |
| Balanced | 1 s | |
| Accurate | 2.4 s | The default: noticeably behind your voice, and the steadiest result |
The delay does not slow down the finish: when you stop, Murmur waits for the tail of the transcription and inserts the whole thing.
Opens the overlay and starts recording, then stops and delivers, the same as Enter. Super+Space by default.
Seconds after which a recording ends on its own and delivers what it has. 600 by default, between 15 and 1800. This is a safety net for a recording you walked away from, not a way to keep dictations short.
Seconds of uninterrupted silence that end the recording. 0 keeps it running until you stop it yourself.
Silence is measured in audio time rather than wall-clock time, so a slow network cannot be mistaken for a pause. Anything quieter than roughly 1% of full scale counts as silence, so a noisy room may need a longer setting, or none.
Not a setting but a live check, with a refresh button. It reports whether Murmur can type with dotool, and names the reason when it cannot: dotool missing, /dev/uinput missing, your user not in the right group, or a group membership that needs a fresh login. See Where the text goes.
Without it Murmur falls back to the shell's virtual keyboard, which reaches every application too but only with characters from your current keyboard layout.
Characters per second, 2500 by default: a sentence lands in a few hundredths of a second. At that speed no key is held at all, so the text goes in as fast as dotool can push it through.
Lower it if characters get dropped or reordered in a particular application, which some Electron and Java applications do under fast synthetic input.
Every setting is a GSettings key under org.gnome.shell.extensions.murmur, which makes them scriptable and easy to keep in dotfiles:
gsettings set org.gnome.shell.extensions.murmur mistral-api-key "$(cat ~/.secrets/mistral)"
gsettings set org.gnome.shell.extensions.murmur toggle-recording "['<Super>space']"
gsettings set org.gnome.shell.extensions.murmur transcription-delay-ms 500
gsettings set org.gnome.shell.extensions.murmur max-recording-seconds 120
gsettings set org.gnome.shell.extensions.murmur silence-timeout-seconds 3
gsettings set org.gnome.shell.extensions.murmur typing-speed 500| Key | Type | Default | Range |
|---|---|---|---|
max-recording-seconds |
integer | 600 | 15 to 1800 |
mistral-api-key |
string | empty | |
silence-timeout-seconds |
integer | 0 | 0 to 30 |
toggle-recording |
string list | ['<Super>space'] |
|
transcription-delay-ms |
integer | 2400 | 240 to 2400 |
typing-speed |
integer | 2500 | 50 to 2500 |
A source install has to point gsettings at the schema it built, since it is not in the system directory:
GSETTINGS_SCHEMA_DIR=$PWD/dist/schemas gsettings get org.gnome.shell.extensions.murmur typing-speed