voculinux/CHANGELOG.md

Changelog

Release history for Vocalinux.

Where to read notes

Source	Contents
GitHub Releases	Canonical release notes per tag
docs/UPDATE.md	Upgrade steps plus retained release notes for recent versions
vocalinux.com/changelog	Shorter website changelog

Current stable

v0.19.1 (2026-11-07)

A patch on 0.19.0. A leftover upstream launcher (vocalinux.desktop, vocalinux, vocalinux-gui) is now reported at startup, because it hid the Voculinux entry in the launcher (#52, #54). The release pipeline docs now say bash for publish-release.sh.

v0.19.0 (2026-11-07)

Release page

The first Voculinux release: a hard fork of 0.18.5 for Void Linux (glibc, x86_64), published as a signed xbps repository from the fork's Gitea host. Changes since v0.18.5, from the commit subjects and PR titles on main:

• Void Linux packaging: signed xbps repository and release pipeline, a package smoke gate across all templates, the voculinux template built from the release tag (#37), and carried templates for python3-pywhispercpp 1.5.0 with CPU variants loaded at runtime (#49), python3-pynput, python3-PyAudio and python3-pysocks, with the audio and SOCKS version floors constrained (#39). The Void package is the only published install channel, and upstream install references are gone (#38, #40).
• Identity: user-visible names are Voculinux, with the identity centralised in one module (tickets #17, #18). An existing configuration and custom dictionary are imported once on first run, and the migrated dictionary points at the fork (#36).
• Model default: a memory-aware whisper.cpp ladder with a conditional backend order (CUDA only with a complete toolkit, then Vulkan, then CPU). Free-memory thresholds are per backend, from the installed-build measurement in #19. A load failure steps down one rung and retries once (ticket #20).
• Dictation on niri: a niri keybind pack and Dank Material Shell notes (ticket #22), a niri-aware inject probe and conflict-free example keybinds (#47). The injection fallback ladder is portal, direct typing, virtual-input tool, XWayland, then clipboard (ticket #31).
• Audio: system audio is captured from PipeWire sink monitors. A configured microphone is never swapped, and the microphone is never opened for system audio (#48).
• Interface and scripting: a dictation state signal and property on D-Bus (ticket #21), a brand accent hook in the GTK colour scheme (ticket #23), and the installed-build performance baseline protocol (ticket #19).
• Website: rebranded to Voculinux (ticket #24), with Google Analytics removed and site links pointing at the fork (#35).
• Fixes: CUDA detection outside PATH, an IBus warning on niri, update checks against the fork's Gitea repository (#36), and the review findings for the Void port (#34).

Carried from upstream after 0.18.5 (VocaHQ/vocalinux PR numbers): overlay Wayland detection (#1030, #1035), quitting when the dock closes the overlay window (#1028, #1036), the selected microphone's full name in the device tooltip (#1029, #1037), escaped device names in the audio status label (#1043), Parakeet Underdog Bark 3.0 models (#1031), the snap build from locked inputs (#1041), the GitHub Actions dependency bump (#1042), and the website's 404 page and SEO work (#1038, #1039, #1040).

See docs/UPDATE.md for the highlight table.

v0.18.5 (2026-10-27)

Patch on the stable line: a stop that races a recording start can no longer leave the microphone recording with no way to halt it, and a stalled microphone read can no longer hang Stop forever, keyboards survive suspend and the GNOME lock screen again because the evdev grab is released when forwarding cannot run, Dictation Pad and Settings no longer bind to other windows or travel workspaces together on Wayland, Ctrl+mouse-wheel zoom works again, the installer gains an upgrade mode that keeps your existing configuration, paste-injection copies are marked sensitive on wl-clipboard 2.2+ so clipboard managers that honor the hint keep them out of history, and AUR source publication is gated on attested release sdists.

See docs/UPDATE.md for the highlight table, or the GitHub Release.

v0.18.4 (2026-10-21)

Patch on the stable line: long push-to-talk dictations are queued in segments instead of being cut off at the buffer cap, Settings never downloads or switches a model without an explicit confirmation and a cancelled download stays cancelled, Test Dictation keeps its own result while the Dictation Pad is open, the GPU picker follows ggml's device list and CPU runs use most of your cores, the recording hotkey cancels a transcription in progress, the Flatpak bundles onnxruntime for Silero VAD and uses the X11 listener on X11 sessions, Wayland injection fixes cover the RemoteDesktop portal, XWayland apps outside IBus and the first recording after launch, Parakeet installs work again, update checks can be turned off, and Slovenian is a supported language.

See docs/UPDATE.md for the highlight table, or the GitHub Release.

v0.18.3 (2026-10-17)

Patch on the stable line: "new line" and "new paragraph" voice commands now insert real line breaks standalone and mid-utterance, hotkeys work again on GNOME Shell X11 and quitting no longer collapses multi-layout keyboards, every packaged channel builds the bundled engine at the portable x86-64 baseline so older CPUs stop crashing, the Snap package drops the uinput plug the Store denied so revisions publish again, the Flatpak remote publish warns loudly instead of skipping silently, and Transcribe File's modal download dialog releases its grab so the transcript's buttons are clickable.

See docs/UPDATE.md for the highlight table, or the GitHub Release.

v0.18.2 (2026-10-12)

Patch on the stable line: the dictation overlay orb now pulses and ripples with live mic level, Nix users get a flake package with NixOS and Home Manager modules plus Vulkan GPU inference for whisper.cpp, held keyboards are left ungrabbed at startup so typing during launch cannot strand a key, the bundled engine no longer crashes on older CPUs, and the piped installer keeps working when it needs to prompt you.

See docs/UPDATE.md for the highlight table, or the GitHub Release.

v0.18.1 (2026-10-08)

Patch on the stable line: a Settings Proxy page for model downloads and update checks behind SOCKS5/HTTP proxies, opt-in JSONL transcript persistence so a failed injection no longer loses the text, .deb and .rpm release packages for Debian and Fedora, a slimmer tray menu (Start on Login and About now live only in Settings), and reliability fixes for the Dictation Pad on Wayland, evdev keyboard discovery, KDE Plasma 6 letter case, a dead Initial Prompt field in Settings, save dialogs that could double the .txt extension, and a Speech Model picker that stages changes until you confirm the download.

See docs/UPDATE.md for the highlight table, or the GitHub Release.

Earlier versions

Browse tags on GitHub Releases. Version history used for maintainer planning also appears in docs/RELEASE_PROCESS.md.

Format

We follow Semantic Versioning. Pre-releases use suffixes such as -alpha, -beta, and -rc.N. Maintainer release steps: docs/RELEASE_PROCESS.md.