v2.48.1 — fixes: tray stop, pinned microphone, Persian dictation, clearer errors¶
Released 2026-11-03. Upgrade:
sudo snap refresh yazses # snap
pipx upgrade yazses # or: uv tool upgrade yazses
scoop update yazses # Windows
Restart the daemon afterwards (yazses restart). This release adds no new feature: every change below corrects something that did not do what it said.
Fixes¶
- The tray's "Stop daemon" stays stopped (Linux). It used to be relaunched about five seconds later. A failed microphone pin, re-calibrate or stop now shows the daemon's own reason instead of a generic message.
- Apply in Settings no longer erases a pinned microphone you did not touch. A pinned device that was unplugged (or a substring pin such as
USB) fell back to "Follow the system default" on screen, and pressing Apply wrote that as a change. - Persian dictation. The built-in English prompt is no longer sent to a non-English decode. The vocabulary corrector no longer rewrites Persian words as if they were English mishearings. The half-space in ordinary Persian no longer triggers a SafeGlyph warning, and the Arabic semicolon splits lines like the ASCII one. The Arabic, Persian and Urdu pages show their headings and tables instead of raw Markdown. On Linux the Unicode injector now resolves keys against the keyboard layout the session reports.
- Clearer failures.
yazses transcribeprints one clear message instead of a traceback and refuses an unknown--modelor a negative speaker count.yazses cite --style bogusexits 2 instead of printing APA, andyazses vocab importnames a binary file or a directory instead of failing obscurely. - Settings window. The focused sidebar row shows a ring, Ctrl+Return applies the staged changes, and labels no longer clip at a large font.
- Biosignal switches (still off by default, still never run on real hardware). Hardening of the quality gate, the stream-integrity guard and the Myo and gForce units, and
yazses doctornow advises on the LSL, ThinkGear, BITalino and Myo sources. - Documentation. The mic-level, first-run, how-to, FAQ and architecture pages were checked against the code and about twenty statements corrected, including eight feature cards that said "Off by default" while first-run setup turns them on.
What was checked: the full test suite, ruff and mypy on the release commit (numbers in the pull request that carries it), and the documentation build.
What was not checked: the tray and the settings window on a real desktop session, any biosignal device, and the Persian changes on recorded audio (no word-error-rate change was measured).
The complete list is in the changelog.