Files

T

xavierk c9be45d3c3 fix(web): link the fork's own source host and drop the upstream star count

The header badge fetched the star count of VocaHQ/vocalinux at build time
and labelled it as Voculinux's. The fetch and the count are gone; the
header, footer and social "source" links now go to the fork repository at
git.bongbetic.com/xavierk/voculinux, as do the issue links, the docs
anchors, the home page JSON-LD and the xbps release card.

Upstream GitHub stays only where it is the real artefact or provenance and
is labelled as such: the install script, the AppImage release card, the
upstream card and the changelog release links. The changelog now says its
entries are upstream history, the open-source page no longer claims
VocaHQ membership or an active GitHub community, and the organization
JSON-LD drops the upstream author as founder. DESIGN.md and PRODUCT.md
follow.

2026-11-06 18:22:17 +05:30

6.6 KiB

Raw Blame History

Product (voculinux.com)

Platform

web (Next.js marketing site under web/)

Users

Linux desktop users who want voice dictation that stays on their machine: developers, writers, and people reducing keyboard strain. They already live in terminals, browsers, IDEs, and GTK/Qt apps.

Product purpose

Voculinux is free, open-source offline voice dictation for Linux. It turns speech into typed text in whatever app has focus, using local engines (whisper.cpp default, Faster Whisper, Whisper, VOSK, Parakeet) or an optional user-configured remote API. Success is: install once, dictate in any app, without cloud transcription or telemetry.

Positioning

Offline-first Linux voice typing with real desktop integration (system tray, X11 and Wayland text injection, toggle and push-to-talk, Silero VAD, suspend recovery). A cloud SaaS voice product cannot truthfully claim local-only models plus no usage telemetry from the installed app.

Operating context

Install via a one-line shell installer, AppImage from GitHub Releases, AUR, Snap (--edge), or Flatpak release bundle; then run from PATH, app menu, or the AppImage binary
System tray indicator and settings GUI (GTK) with searchable sidebar navigation and an always-visible sidebar footer for dictation status / Test Dictation / Close
Dictation into terminals, browsers, IDEs, office apps
Engines and models chosen for hardware (CPU, optional Vulkan GPU with discrete-device preference)
Guides and comparison pages on voculinux.com; source on the fork's own Gitea host (https://git.bongbetic.com/xavierk/voculinux); upstream GitHub is linked only as labelled provenance

Capabilities and constraints

Engines: whisper.cpp (default), Faster Whisper, OpenAI Whisper, VOSK, Parakeet, optional Remote API
Speech languages: large selectable catalog (~33 + Auto-detect) shared by Settings/CLI; VOSK only lists languages with official Alphacephei models; remaining Whisper languages available via Auto-detect
Display servers: X11 and Wayland
Shortcut modes: push-to-talk default (hold Right Alt / Option); toggle available; left/right modifier distinction; configurable modifier+key combos
Searchable language combobox; delete unused downloaded speech models from Settings
Optional voice commands with localized punctuation phrases for common languages; Silero neural VAD with amplitude fallback
Dictation Pad: in-app window that receives dictation for manual copy-out; the Wayland-safe path that skips text injection entirely
Per-language dictation shortcuts, optional follow of the active keyboard layout, and a tray history menu with recent dictations
Floating dictation overlay, optional lowering of other audio while dictating, and custom dictionary terms bias plus transcript corrections
Audio capture via PipeWire (microphone and system-audio sources) or PortAudio; Wayland text injection via IBus, wtype, ydotool, or the RemoteDesktop portal
Opt-in D-Bus activation so compositor global shortcuts and scripts can start dictation
File transcription with per-speaker labels (--transcribe-file / tray), powered by TinyDiarize
Continuous dictation polish: capitalize after sentence punctuation; trailing space after each completed utterance
Optional auto-pause while configured apps run; optional idle model keep-alive unload
Optional JSONL persistence of transcription history to disk (off by default)
Settings Proxy page: outbound SOCKS5 or HTTP CONNECT proxy (optional auth) for model downloads and update checks
Speech Model Advanced pickers stage engine/size/variant/language changes until Download is confirmed
In-app update checker (stable/nightly) with tray notification when a newer GitHub release is available
Settings About page groups this app, upstream project links retained as provenance, and talk-to-us links (GitHub, Discord, X, email)
Optional disable of the missing-tray warning dialog
Settings → Audio tone picker (Lift, Flick, Ember, Step, Voca, Soft, Chirp, Scale, Drop, Glass, Off) with Preview; new installs default to Voca; catalog files are the preview WAVs
Vulkan discrete GPU auto-select with manual device override in Advanced settings
Native GTK installs follow the OS dark/light preference unless GTK_THEME is already set
Wayland: IBus when ibus-wayland is running, including on compositors previously treated as unbridged
Packaging: install.sh (distro python3-gi required; no pip sdist of PyGObject), .deb/.rpm (x86_64/aarch64), AppImage (x86_64/aarch64), AUR, Flatpak (local/Flathub status as documented); uv.lock pins Python deps; Justfile replaces Makefile
No usage telemetry in the installed app
AGPL-3.0; marketing version string is tracked in site package/version surfaces
Website is Next.js marketing + SEO guides (static export); languages page documents per-engine support honestly

Brand commitments

Name: Voculinux, a hard fork of Vocalinux for Void Linux (glibc, Wayland-primary under niri). Upstream project links are retained as provenance; the fork site carries the new name everywhere a user looks
Mark: fork mic for the app icon, tray states, and site favicons (public/); Bongbetic appears only as attribution chrome in the site header and footer, linking the hosted brand assets at https://bongbetic.com/branding/ — no brand binaries are vendored into the repository
Visual system: Carbon-mapped — Carbon spacing, type-role, motion, and contrast discipline remapped to the house palette (warm paper canvas, teal interactive colour, system type stacks only, red reserved for recording/destructive); no Carbon blue, no remote webfonts
Engine promise: whisper.cpp is the guaranteed default engine (packaged, checksum-pinned, budgeted); Faster Whisper, Whisper, VOSK, and Parakeet are best-effort installable extras; the Remote API engine is user-configured and audio leaves the machine — stated plainly wherever remote is described
Voice: practical, specific, Linux-native; not hype-first SaaS copy

Evidence on hand

Real app screenshots in public/screenshots/
Install/uninstall commands pointing at GitHub raw install scripts
JSON-LD on the home page (software application, FAQ, how-to)
Do not invent user counts, testimonials, or benchmarks

Product principles

Privacy is default: local engines process audio on-device; remote only when the user opts in.
Ship the install path and real desktop proof, not abstract feature theater.
Stay honest about engines, hardware, and what runs offline vs remote.
Prefer Linux craft and clarity over generic AI-startup marketing patterns.

Accessibility

Respect prefers-reduced-motion
Body text contrast at least 4.5:1; large text at least 3:1
Keyboard-reachable install copy controls and navigation

6.6 KiB Raw Blame History