blitztext-app-linux/linux/CHANGELOG.md

Changelog

All notable changes to Blitztext for Linux (the native dictation tool in linux/) are documented here. The format follows Keep a Changelog, and this project adheres to Semantic Versioning.

The version is defined in blitztext/__init__.py.

Unreleased

[1.8.0] - 2026-06-08

Added

• Send by voice: say a distinctive phrase like "computer send" at the start or end of a clip and the word is stripped, then the rest is typed and submitted with Enter — the spoken equivalent of "stop + paste + Enter". Mainly for hands-free use, where you can't press a key. Configure under Settings → Input → "Send words", or [routing] send_keywords. Off by default; because it presses Enter, use a multi-word phrase (e.g. your wakeword + "send") so a sentence that merely ends in "send" doesn't submit by accident. Matched the same edge-anchored, ASR-tolerant way as routing/cancel keywords.
• Wakeword benchmark (Settings → Benchmark): stress-test hands-free detection. It synthesizes short sentences with your wake phrase spoken in random voices (plus pure-filler utterances with none), streams them to your wyoming-openwakeword server, and reports recall (how reliably it fires), false fires, and a per-voice breakdown. Speech comes from any OpenAI-compatible TTS server (Kokoro-FastAPI, XTTS, OpenAI, …): set its URL, optional API-key env var, model, and voices under the new [tts] config / the Benchmark tab, and use Connect to test it (it auto-fills the voice list when the server exposes one).

[1.7.1] - 2026-06-08

Fixed

• Overlay waveform and silence countdown ring never appeared on systems where PortAudio/sounddevice can't open the default input — notably PipeWire boxes, where opening an input stream simply hangs. Both the live waveform and the auto-stop countdown are driven by a single level meter, which was the only part of the app still using sounddevice (everything else records via pw-record). The meter now streams raw PCM from the same system recorder as the WAV recorder (pw-record/parecord/arecord) and computes the level itself, so it works wherever recording works — on both the hotkey and hands-free (wakeword) paths, plus the mic-level preview in Settings. No more PortAudio dependency for metering.
• App reported itself as "__main__.py" in the taskbar and in GNOME's "… is not responding" dialog. Launched via python -m blitztext, GTK's default program name is argv[0]'s basename. It now sets prgname/application name to Blitztext before any window is realized (and the desktop file gains StartupWMClass=blitztext for the .desktop match + icon), without touching the -m blitztext entry point.

[1.7.0] - 2026-06-07

Added

• Spoken cancel keyword: say a word like "abbrechen" (or "cancel") at the start or end of a clip and the whole dictation is discarded — it is never routed, rewritten, or typed anywhere. Mainly rescues an accidentally triggered (e.g. wakeword) recording. Configure under Settings → Mic/Cues → "Cancel words", or [routing] cancel_keywords (default ["abbrechen", "cancel"]; empty list disables it). Matched the same edge-anchored, ASR-tolerant way as routing keywords, so the word buried mid-sentence won't trip it.

[1.6.0] - 2026-06-07

Fixed

• Session freeze when the overlay's caret tracking was active (could lock up the whole GNOME/X11 desktop, forcing a logout/reboot). The AT-SPI caret tracker subscribed to the high-frequency object:text-caret-moved signal and made synchronous, blocking AT-SPI reads from inside the event handler — which re-enters the accessibility dispatcher and is stormed by the app's own xdotool typing (one event per character), congesting the a11y bus until the desktop stopped responding. It now tracks focus changes only and reads the caret rectangle lazily (once, when the overlay shows), never from inside an event dispatch.

Changed

• Matched preset is fused into the overlay instead of a desktop notification: when voice routing picks a preset, the overlay shows its emoji icon, name, and the spoken keyword on a banner, and narrates the phase ("Transcribing…" → "Rewriting…"). With the overlay on, the redundant per-dictation notifications are suppressed (errors still notify); headless/overlay-off keeps notifications.

Added

• Live LLM rewrite in the overlay: rewrite presets now stream the model's output into the bubble token-by-token, so you watch it write. The delivered text is still the complete result, typed once the rewrite finishes.

1.5.1 - 2026-06-07

Added

• Silence auto-stop countdown ring on the dictation overlay: when you stop speaking, a full circle wrapping the microphone glyph drains clockwise as the trailing-silence timer runs out, recolouring from calm cyan to an urgent red and emptying exactly as the recording auto-stops. It spans the configured "Silence to stop (s)" window ([wakeword] silence_seconds), fades back in/out as you pause and resume, and so finally makes the hands-free auto-stop visible instead of a silent surprise.

1.5.0 - 2026-06-07

Added

• On-screen dictation overlay (Settings → General → "Visual overlay", or [general] overlay_enabled, default on): the moment recording starts — by hotkey or wakeword — a translucent bubble appears at the cursor showing a pulsing microphone, a live waveform of your mic level, and the recognised text (word-by-word with a realtime streaming STT engine, or the final result as a brief confirmation otherwise). Its tail points at where the text will land: it follows the text caret when the focused app exposes it over accessibility (AT-SPI), otherwise the mouse pointer, otherwise a screen corner — tune via [general] overlay_anchor = "caret" | "pointer" | "corner". The window is click-through and never takes focus, and it finally gives hands-free wakeword sessions visible feedback (their notifications are suppressed by design). X11 only; falls back to a corner where the cursor can't be located.

Changed

• Presets are speakable by name: voice routing now matches a preset's name as an implicit keyword, so a preset works by voice even with no keywords configured (e.g. just say "nicer email …"). Explicit keywords still take precedence, and preset names also bias the STT for better recognition.
• General settings switches moved to the far right of each row, each with an inline description so it's clear what the toggle does without hovering.
• About: added a "Copyright: 2026 mARTin Bierschenk - Design" line.

[1.4.0] - 2026-06-07

Added

• "Announce matched preset" notification (Settings → General, or [general] notify_routing, default on): after a voice command, a notification shows which preset and spoken keyword matched — shown even for hands-free wakeword sessions, so you can see what you triggered. It only fires on a real match, so it never spams when nothing is said.
• Per-preset emoji icon (Presets → "Icon (emoji)"): give each preset a distinct emoji, shown in the matched-preset notification so you can tell at a glance which fired.

Fixed

• Voice-routing default went to a rewrite: when no [routing] default preset is set, the no-keyword fallback used the first preset — which, if that happened to be an LLM rewrite (e.g. "Improve text"), sent every unrouted wakeword command to the language model (and failed when the LLM was down). The fallback now prefers a transcribe preset, so the default action is plain transcription.

[1.3.0] - 2026-06-07

Added

• Pause wakeword (tray): a reversible "Pause wakeword" toggle appears in the system-tray menu when the wakeword is enabled. It pauses/resumes hands-free detection by toggling the /tmp/wake_muted flag (external scripts may toggle the same file).
• "Play audio cues" switch (Settings → Input → Audio cues, or [sounds] enabled): on/off for the manual (keyboard/hotkey) start/stop chimes. Defaults to on. The hands-free wakeword sounds are independent of it.
• Configurable wakeword auto-stop silence (Settings → Input → Hands-free → "Silence to stop (s)", or [wakeword] silence_seconds): end a hands-free recording this many seconds after you stop speaking. Defaults to 2.0 (previously hard-coded to 2.5 s).

Fixed

• Wakeword sounds silenced by the manual cue switch: the "Play audio cues" master switch wrongly muted the hands-free Sound: detected/captured cues too. Wakeword cues are now independent — they play whenever a file is set and stay silent when cleared (no surprise system-chime fallback), regardless of the manual switch.
• PortAudio/ALSA teardown noise: the level meter no longer leaks pthread_join ... failed / PaUnixThread_Terminate ... failed lines to the terminal when a clip ends — that C-library chatter (written straight to fd 2) is now suppressed around the stream open/close.
• Wakeword stuck muted: a leftover /tmp/wake_muted flag silently disabled detection with no in-app way to clear it. The state is now exposed and reversible from the tray, so a stale flag no longer kills hands-free use. The daemon also logs a clear Starting PAUSED warning when it boots muted.
• Away-from-keyboard "Busy" storm: a wakeword hit arriving while the previous clip was still transcribing went through toggle() and popped a "Busy" notification. Wakeword triggers now go straight to start_dictation(), so a busy/not-ready state is ignored silently instead.
• Quiet hands-free errors: transcription/rewrite failures during a wakeword-triggered session no longer raise critical desktop notifications — they are logged instead, keeping background sessions silent.
• Notification storm / lock-screen pile-up: desktop notifications are now sent as transient with a short expiry and reuse a single bubble, so they no longer stack in the notification log or persist on the lock screen.
• Quiet hands-free sessions: per-dictation notifications are suppressed for wakeword-triggered sessions (audio cues are used instead).

[1.2.0] - 2026-06-05

Added

• Wyoming Wakeword Support: Complete hands-free integration via Wyoming protocol (e.g., openWakeWord), with live configuration testing and model fetching in the UI.
• ATK Screen Reader Accessibility: Fully mapped GTK labels, inputs, tooltips, and properties to the ATK bridge, enabling seamless navigation for blind users via screen readers like Orca.
• Drag-and-Drop Workflow Ordering: Workflows in the main tray menu can now be reordered via native drag-and-drop.
• Voice Activity Detection (VAD) Auto-Stop: Dictation now automatically stops after detecting 2.5 seconds of silence, removing the need to manually click Stop.
• Audio Feedback: Added audible start/stop/cancel chimes mapping to system-native alert sounds.
• Benchmark Autocomplete: The Benchmark UI automatically fills in the reference .txt transcript if it matches the selected audio file.
• Realtime STT streaming mode: new mode = "stream" workflow support and a riva_realtime STT engine for Riva/NIM WebSocket transcription, including a Settings shortcut for Nemotron ASR Streaming on http://127.0.0.1:8006/v1.
• Settings About tab with the app version, source link, changelog, and license text.
• STT & LLM engine manager: add, rename, and delete engine presets, each with an online/offline status dot, a per-engine type (local/cloud), and a searchable model dropdown populated from the server's /models (with a reload button). Local Whisper device/precision now live with the STT engine.
• Benchmark tab: run a reference clip through every STT engine and compare time, case-sensitive accuracy (WER), and a CPU/GPU/remote device column, with the fastest and most accurate highlighted.
• Custom audio cues: pick your own WAV files to play when recording starts and stops (covers stop+paste, stop+paste+Enter, and silence auto-stop), each with play-test and clear-to-default buttons. Built-in system sounds otherwise.
• Settings Log tab: a live activity log (model load/download, transcriptions, errors) with Copy and Clear, so a long "Loading…" is no longer opaque.
• Per-tab info boxes and expanded tooltips across Settings, written in plain language and exposed to screen readers (ATK) — for non-technical and blind users (barrierefrei).
• Click-to-bind hotkeys: a Set button captures the next keypress (including modifier-only chords like Ctrl+Win) into any hotkey field.

Changed

• GUI rebuilt in GTK 3 (replacing tkinter): a native GNOME panel unified with the tray, with the Ubuntu font and a dropdown+editor pattern in Settings.
• Voice-keyword routing and a modifier hotkey scheme (Ctrl+Win start, Ctrl stop+paste, Alt stop+paste+Enter, Esc cancel) replace per-preset combos as the default way to dictate.
• Quality gate rejects silent/too-short clips and stock Whisper hallucinations before they reach the screen.

1.1.0 - 2026-06-04

Added

• Debian package (packaging/build-deb.sh) producing an installable blitztext_<ver>_arm64.deb with a desktop entry, app icon, and a blitztext launcher. Installs via the Software app or apt install ./…deb. Bundles a relocatable venv with all Python deps (no pip/network at install) and declares system deps (python3-gi, xdotool, libnotify-bin, a recorder) so they pull in automatically. The bundled venv is built on the system /usr/bin/python3, so the tray works out of the box.

Notes

• python3-gi is already present on a standard Ubuntu GNOME install; the tray only seemed unavailable from source when the project venv was built from a non-system Python (e.g. conda/miniforge). The .deb avoids this entirely.

1.0.1 - 2026-06-03

Changed

• Redesigned the control-panel window: minimal flat layout, Ubuntu font throughout, clickable workflow rows with hover (click to record / stop), subtle dividers, and text-style Settings/Quit actions. Dropped the monogram avatars and per-row buttons in favour of a cleaner, simpler look. The Settings window picks up the same font and styling.

1.0.0 - 2026-06-03

First release of the Linux port. The upstream project is a macOS-only menu-bar app (Swift/SwiftUI, CoreML/WhisperKit) that cannot run on Linux or in a container; this is a native host tool that reproduces the workflow — focus any text field, press a hotkey, speak, and the (optionally rewritten) text is typed into that field.

Added

• Native dictation engine (daemon.py): global hotkeys via pynput, each hotkey toggles record → transcribe → optional rewrite → deliver.
• Local transcription via faster-whisper (transcribe.py); device="auto" tries CUDA and falls back to CPU int8 (CPU-only on this arm64 host).
• Microphone recording (recorder.py) through pw-record / parecord / arecord — 16 kHz mono WAV, no Python audio bindings required.
• Optional LLM rewrite (rewrite.py) against any OpenAI-compatible endpoint (OpenAI, or a local vLLM / llama-swap), configurable per workflow.
• Typing into the focused window via xdotool (paste.py): type directly or paste through the clipboard; re-activates the target window first.
• Configurable workflows (config.py) in ~/.config/blitztext/config.toml, with five defaults (Transcribe, Nicer email, Improve text, Calm down, Add emojis), per-workflow prompt/model/temperature overrides, and a TOML writer.
• Control-panel window (gui.py, tkinter): workflow rows with monogram avatars, hotkey badges, per-row Record buttons, a live status dot, and a Settings window that edits config and can Save & Restart.
• System-tray mode (tray.py, AppIndicator): macOS-menu-bar-style status icon with a menu to trigger each workflow, Show panel, Settings…, and Quit; shares one daemon/model/hotkey set with the window. Falls back to the window with an install hint when PyGObject is absent.
• CLI (__main__.py): tray (default), gui, run, transcribe, config-path, and --version.
• Packaging: install.sh (venv with --system-site-packages), requirements.txt, and a blitztext.service systemd user unit.

Notes

• Targets an X11 session (uses xdotool); Wayland would need ydotool/wtype.
• System tray requires a one-time sudo apt install python3-gi (the GTK / AppIndicator typelibs and GNOME ubuntu-appindicators extension are already present on the target host).