v0.13.0-beta

This release is all about making dictation feel solid on Wayland, plus a much smarter whisper.cpp model picker. If you're on a wlroots-style compositor (COSMIC, sway, Hyprland, niri), a non-US keyboard layout, or you've ever watched text silently fail to appear, this one is for you.

A lot of what's here came straight from people filing clear bug reports and sending fixes. Thank you. 🙏

✨ New features

• Guided whisper.cpp model selection. The Settings dialog now splits whisper.cpp into a Model Size and a Specialization picker, so you can choose English-only, quantized (Q5/Q8), Large v3 Turbo, or legacy large variants, with language-aware recommendations and hover hints. You can also pass exact model IDs like medium.en-q5_0 or large-v3-turbo straight to --model. (#465, closes #461, thanks @rafaeln)

🐛 Reliability fixes (mostly Wayland)

• Silent text drops on wlroots/COSMIC compositors. Vocalinux now uses wtype/ydotool instead of IBus on compositors that don't bridge IBus to native apps (COSMIC, sway, Hyprland, and similar), so injection stops silently doing nothing. (#486, closes #485, thanks @bbbboother)
• IBus false-positive that swallowed text. On Wayland we now require a real IM engine before trusting IBus, so a bare xkb:* layout (including on GNOME/Mutter) no longer reports success while inserting nothing. (#491, closes #478, thanks @subsy)
• Keyboard layout flipping to us after dictation. We no longer run setxkbmap on Wayland, where it was quietly rewriting the XWayland layout and leaving XWayland/Electron apps on the wrong keys. (#490, closes #474, thanks @bcs-23, and @fsioni and @karottenreibe for tracing the cause)
• Garbled output on non-US layouts (AZERTY/QWERTZ/Dvorak) and a clipboard-copy hang. ydotool now pastes through the clipboard, which is layout-independent and Unicode-safe. (#480, thanks @fsioni)
• Faster ydotool typing. An explicit --key-delay noticeably cuts injection latency. (#488, closes #482, thanks @subsy)
• Hotplugged keyboards now work. The evdev shortcut backend rescans, so a keyboard plugged in after startup is picked up and reconnects recover. (#467, closes #431, thanks @LuigiKraken)
• Dictation spacing preserved. Spacing between speech segments separated by a pause is kept, so words no longer run together after a silence. (#464)
• KDE Plasma Wayland guidance. Vocalinux detects KDE Plasma Wayland and points you to enable IBus Wayland when wtype injection fails, both in the installer and in the logs. (#466, closes #463, thanks @cyb3rko)

🔧 Under the hood

• Refreshed website docs with dedicated pages for the Remote API engine, Silero VAD, advanced whisper.cpp settings, and desktop reliability. (#470)
• Automatic PR labeling by changed files. (#473)

🚀 Try it

New install or update (recommended, it handles system deps, models, and desktop integration):

curl -fsSL https://raw.githubusercontent.com/jatinkrmalik/vocalinux/main/install.sh | bash

From source:

git clone https://github.com/jatinkrmalik/vocalinux.git
cd vocalinux && ./install.sh

Already running a source checkout:

cd vocalinux && git fetch origin && git checkout v0.13.0-beta && ./install.sh

To try the new model picker, open Settings, go to Speech Engine, and choose a Model Size plus a Specialization. Or from the CLI: vocalinux --model large-v3-turbo.

💚 Thank you

Vocalinux keeps getting better because of the people who use it and push back on it. Big thanks to the contributors who sent code this cycle, @fsioni and @bbbboother, and to everyone whose reports and feature requests turned into these fixes: @rafaeln, @cyb3rko, @LuigiKraken, @subsy, @bcs-23, and @karottenreibe. This is a project I rely on every single day, and it genuinely wouldn't be this good without you.

If something breaks, or you have an idea for what would make Vocalinux better, please open an issue or start a discussion. Feedback and suggestions are always welcome.

Full changelog: v0.12.0-beta...v0.13.0-beta