The daemon now requests the Bluetooth headset profile itself (pactl
set-card-profile) the moment a take starts, instead of waiting for
WirePlumber to notice the stream and run its switch timer. The microphone
link comes up 0.15-0.18 s after the key instead of 0.21-0.27 s; WirePlumber
still restores A2DP when the stream goes away. First real audio is now at
0.82-0.86 s, which is the headset's own floor.
When the plugin is updated while the daemon runs, the daemon sees its source
change, exits with code 4 once idle, and the shell service relaunches it
without counting that as a crash.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Xoc9DJCViR7dg9eKzzAcfQ
The bar turned green after 2 s on a timer even when the microphone delivered
nothing. It now turns green only on real audio: RMS above 0.0005, or three
consecutive chunks that are not digital silence (the virtual mic emits one
stray nonzero chunk right after start, and a Bluetooth headset's floor ramps
in from a few LSB). Measured 1.2-1.4 s after the key, within 0.14 s of the
first real samples.
diagnostics/bluetooth/ records how startup went from ~1.6 s (or never) to
~1.1-1.4 s: a btusb driver bug, two PipeWire bluez5 bugs, a 500 ms WirePlumber
switch timeout and an over-broad auto-connect rule. Those fixes are machine
level and live outside this repo; the patches, tools and measurements are here.
STARTUP-TIME.md is the summary and explains the ~0.9 s hardware floor.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Xoc9DJCViR7dg9eKzzAcfQ
- A language can be added several times, each entry with its own key,
+ Return switch and agent key (ids en, en-2, …; the keybindings list
shows 'English 2, sends'). Rows show 'English · 2'.
- Advanced → Keep mic open: off / 2 minutes after a recording / 10 minutes /
always. The timed options give instant starts during a burst of dictation
while letting a Bluetooth headset return to its music profile afterwards.
- The bar speaks the language being dictated ("Ouvindo…", "Escuchando…", …).
- Green now means real audio is arriving, not just that pw-record connected:
a Bluetooth microphone sends silence while it switches profile, which is
where the first words were being lost.
- Advanced → "Keep the microphone open": the stream is parked between
recordings and a recording starts instantly with ~0.6 s of pre-roll.
- Settings → "Bar animation": bars, wave, pulse or dots.
- pw-record is shut down off the event loop so a cancel can never stall.
- Settings: "Keep history" (a day … forever, default a month); older
recordings and their audio are pruned at startup, after each recording
and when the setting changes.
- History rows keep play, copy and delete only.
- Advanced: the microphone is picked from PipeWire's sources instead of a
free-text field, with a note about Bluetooth headsets switching profile.
- A stock machine without voxtype gets a banner with an Install button that
runs omarchy-voxtype-install in a floating terminal; the daemon refuses to
record until the tools are there and says why.
A bar widget plus a stdlib-only Python daemon. Press a language's key to
record, press it again to stop: the text is pasted at the cursor (or handed
to the default coding agent with a second key). While recording the bar
shows a waveform (yellow while the microphone opens, green while listening)
and the words as they are recognised; the recording is transcribed at every
pause, so stopping only transcribes the last phrase. Every recording and its
text are kept in a searchable history with playback, copy, paste and delete.
Uses Omarchy's own dictation engine (voxtype, local Whisper) by default and
downloads any model a language needs by itself; whisper.cpp or a custom
command can be picked instead. Key bindings are applied at runtime through
Hyprland's Lua API and never touch a key something else already uses.
alan