scribe-cli 1.1.0__tar.gz → 1.1.1__tar.gz

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scribe-cli
-Version: 1.1.0
+Version: 1.1.1
 Summary: Speech-to-text CLI and system-tray app for dictating into any focused window. Local (vosk, faster-whisper) or cloud (groq, openai) backends, batch or streaming.
 Author-email: Mahé Perrette <mahe.perrette@gmail.com>
 License: MIT License
@@ -72,8 +72,10 @@ Requires-Dist: pyperclip
 Requires-Dist: unidecode
 Requires-Dist: termcolor
 Requires-Dist: platformdirs
-Requires-Dist: desktop-ai-core>=0.2.0
+Requires-Dist: desktop-ai-core>=0.3.1
 Requires-Dist: onnxruntime
+Requires-Dist: pynput
+Requires-Dist: pystray
 Provides-Extra: keyboard
 Requires-Dist: pynput; extra == "keyboard"
 Provides-Extra: whisper
@@ -83,8 +85,7 @@ Requires-Dist: pywhispercpp; extra == "whisper-futo"
 Provides-Extra: vosk
 Requires-Dist: vosk; extra == "vosk"
 Provides-Extra: app
-Requires-Dist: pystray; extra == "app"
-Requires-Dist: PyGObject; extra == "app"
+Requires-Dist: PyGObject; sys_platform == "linux" and extra == "app"
 Provides-Extra: openai
 Requires-Dist: openai<3,>=2.37.0; extra == "openai"
 Requires-Dist: soundfile; extra == "openai"
@@ -93,13 +94,11 @@ Requires-Dist: openai<3,>=2.37.0; extra == "groq"
 Requires-Dist: soundfile; extra == "groq"
 Provides-Extra: vad
 Provides-Extra: all
-Requires-Dist: pynput; extra == "all"
 Requires-Dist: faster-whisper; extra == "all"
-Requires-Dist: pywhispercpp; extra == "all"
 Requires-Dist: openai<3,>=2.37.0; extra == "all"
 Requires-Dist: soundfile; extra == "all"
 Requires-Dist: vosk; extra == "all"
-Requires-Dist: pystray; extra == "all"
+Requires-Dist: PyGObject; sys_platform == "linux" and extra == "all"
 Dynamic: license-file
 
 [![pypi](https://img.shields.io/pypi/v/scribe-cli)](https://pypi.org/project/scribe-cli)
@@ -129,13 +128,29 @@ cloud-based APIs, batch and streaming workflows.
 
 ## Install
 
+**Linux / macOS:**
+
 ```bash
 sudo apt-get install portaudio19-dev xclip   # Ubuntu; macOS: brew install portaudio
 pip install scribe-cli[all]
 export GROQ_API_KEY=YOURAPIKEY                # or OPENAI_API_KEY, or skip and run local
 ```
 
-See documentation below for setting up keyboard input on Ubuntu Wayland.
+**Windows** (PowerShell) — no system packages needed; `sounddevice` bundles
+PortAudio and the clipboard is native, so skip the `apt`/`brew` step:
+
+```powershell
+py -m venv .venv
+.\.venv\Scripts\Activate.ps1          # if blocked: Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
+pip install scribe-cli[whisper]        # local Whisper; or [all], or a cloud backend
+$env:GROQ_API_KEY = "YOURAPIKEY"        # or OPENAI_API_KEY, or skip and run local
+```
+
+The tray app and keyboard typing work out of the box on Windows — `pynput`
+and `pystray` are regular dependencies, and there is nothing to create by
+hand (no `C:\tmp`). See [docs/installation.md](docs/installation.md#windows)
+for the full Windows walkthrough, and the documentation below for setting up
+keyboard input on Ubuntu Wayland.
 
 
 ## Usage
@@ -237,8 +252,24 @@ I personally use [OpenAI](https://openai.com/api/) with `gpt-4o-mini-transcribe`
 
 ## Compatibility
 
-Initially developed for Python 3 on Ubuntu 24.04 (GNOME + Wayland);
-works on macOS and Windows too. Wayland keystroke injection is
-convoluted but [solved](docs/output.md). For dependencies of
-individual subsystems, check `pynput` (keyboard) and `pystray` (tray
-icon).
+| OS                 | Status                                                              |
+|--------------------|---------------------------------------------------------------------|
+| Ubuntu 24.04       | Primary dev platform (GNOME, X11 and Wayland).                      |
+| macOS              | Works.                                                              |
+| Windows 11         | Tested and working on Python 3.14 (64-bit / `win_amd64`). Every dependency resolves a ready-made wheel — no toolchain or Python downgrade needed. |
+
+Wayland keystroke injection is convoluted but [solved](docs/output.md).
+For dependencies of individual subsystems, check `pynput` (keyboard) and
+`pystray` (tray icon).
+
+**Windows notes:**
+
+- The tray icon is hidden under the taskbar overflow arrow (`^`) by
+  default. Pin it via *Settings → Personalization → Taskbar → Other
+  system tray icons*.
+- A **single click** on the tray icon fires the default action (Record).
+  This is a free bonus of pystray's Win32 backend; on Ubuntu the
+  AppIndicator backend only opens the menu (a backend limitation, not a
+  bug).
+- If recording fails, allow mic access under *Settings → Privacy &
+  security → Microphone → "Let desktop apps access your microphone"*.

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/README.md

@@ -25,13 +25,29 @@ cloud-based APIs, batch and streaming workflows.
 
 ## Install
 
+**Linux / macOS:**
+
 ```bash
 sudo apt-get install portaudio19-dev xclip   # Ubuntu; macOS: brew install portaudio
 pip install scribe-cli[all]
 export GROQ_API_KEY=YOURAPIKEY                # or OPENAI_API_KEY, or skip and run local
 ```
 
-See documentation below for setting up keyboard input on Ubuntu Wayland.
+**Windows** (PowerShell) — no system packages needed; `sounddevice` bundles
+PortAudio and the clipboard is native, so skip the `apt`/`brew` step:
+
+```powershell
+py -m venv .venv
+.\.venv\Scripts\Activate.ps1          # if blocked: Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
+pip install scribe-cli[whisper]        # local Whisper; or [all], or a cloud backend
+$env:GROQ_API_KEY = "YOURAPIKEY"        # or OPENAI_API_KEY, or skip and run local
+```
+
+The tray app and keyboard typing work out of the box on Windows — `pynput`
+and `pystray` are regular dependencies, and there is nothing to create by
+hand (no `C:\tmp`). See [docs/installation.md](docs/installation.md#windows)
+for the full Windows walkthrough, and the documentation below for setting up
+keyboard input on Ubuntu Wayland.
 
 
 ## Usage
@@ -133,8 +149,24 @@ I personally use [OpenAI](https://openai.com/api/) with `gpt-4o-mini-transcribe`
 
 ## Compatibility
 
-Initially developed for Python 3 on Ubuntu 24.04 (GNOME + Wayland);
-works on macOS and Windows too. Wayland keystroke injection is
-convoluted but [solved](docs/output.md). For dependencies of
-individual subsystems, check `pynput` (keyboard) and `pystray` (tray
-icon).
+| OS                 | Status                                                              |
+|--------------------|---------------------------------------------------------------------|
+| Ubuntu 24.04       | Primary dev platform (GNOME, X11 and Wayland).                      |
+| macOS              | Works.                                                              |
+| Windows 11         | Tested and working on Python 3.14 (64-bit / `win_amd64`). Every dependency resolves a ready-made wheel — no toolchain or Python downgrade needed. |
+
+Wayland keystroke injection is convoluted but [solved](docs/output.md).
+For dependencies of individual subsystems, check `pynput` (keyboard) and
+`pystray` (tray icon).
+
+**Windows notes:**
+
+- The tray icon is hidden under the taskbar overflow arrow (`^`) by
+  default. Pin it via *Settings → Personalization → Taskbar → Other
+  system tray icons*.
+- A **single click** on the tray icon fires the default action (Record).
+  This is a free bonus of pystray's Win32 backend; on Ubuntu the
+  AppIndicator backend only opens the menu (a backend limitation, not a
+  bug).
+- If recording fails, allow mic access under *Settings → Privacy &
+  security → Microphone → "Let desktop apps access your microphone"*.

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/docs/backends.md

@@ -275,20 +275,13 @@ Once the buffer has grown to at least `--stream-chunk-min` (default
 (default 10 s) regardless of silence, to cap latency. The session
 continues until you stop it manually.
 
-The **first** chunk of a streaming thread uses a different floor:
-`--stream-first-chunk-min` (default 3 s). The bootstrap chunk has no
-prior text to bias Whisper's punctuation/casing, so a longer audio
-window lets the model produce a properly-punctuated transcript whose
-tail then seeds the rolling prompt for every chunk after it.
-Subsequent chunks fall back to `--stream-chunk-min`. The override
-also re-engages right after a context-reset silence (i.e. when a long
-pause cleared the rolling tail — see *Cross-chunk prompt context*
-below). Set `--stream-first-chunk-min` equal to `--stream-chunk-min`
-to disable the override. It's automatically inactive when
-`--stream-context-length 0` (Patient profile), where there is no
-rolling context to bootstrap. Internally clamped to `≤
---stream-chunk-max` so a misconfigured pair can't deadlock the
-chunker.
+The first chunk uses a higher floor (`--stream-first-chunk-min`,
+default 3 s) so the bootstrap chunk has enough audio to seed the
+rolling prompt for the rest. Auto-disabled when
+`--stream-context-length 0` (Patient). If you stop talking before
+the floor is reached, a pause past `--stream-context-reset-silence ×
+--stream-chunk-silence-break` (default 1.8 s) flushes the buffer
+anyway — your utterance is never stranded.
 
 ### Does pseudo-streaming change the API cost?
 

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/docs/installation.md

@@ -21,7 +21,10 @@ On macOS use Homebrew:
 brew install portaudio
 ```
 
-(Windows ships everything needed via the wheels.)
+On Windows there are **no system packages to install**: `sounddevice`
+bundles PortAudio in its wheel and the clipboard uses the native Windows
+API, so neither `portaudio19-dev` nor `xclip` apply. See the
+[Windows quickstart](#windows) below.
 
 ## Python package
 
@@ -50,14 +53,52 @@ the four backends and the tray UI:
 | `[vosk]`     | `vosk`                                             | local Vosk backend (streaming)          |
 | `[openai]`   | `openai`, `soundfile`                              | OpenAI cloud backend (incl. realtime)   |
 | `[groq]`     | `openai`, `soundfile`                              | Groq cloud backend                      |
-| `[keyboard]` | `pynput`                                           | the `pynput` typer (XTest/Quartz/WinAPI)|
-| `[app]`      | `pystray`, `PyGObject`                             | system tray icon                        |
-| `[all]`      | all of the above                                   | one-shot setup                          |
+| `[keyboard]` | `pynput`                                           | back-compat only — `pynput` is a base dep now |
+| `[app]`      | `PyGObject` (Linux only)                            | the Linux AppIndicator tray binding     |
+| `[all]`      | every backend + Linux tray binding                 | one-shot setup                          |
+
+> **`pynput` and `pystray` are base dependencies.** The default run uses
+> the keyboard typer and the system-tray app, so both ship with the plain
+> `pip install scribe-cli` — you do **not** need `[keyboard]` or `[app]`
+> for the standard experience. `[app]` now only adds the Linux-only
+> `PyGObject` AppIndicator binding (skipped automatically on Windows/macOS
+> via a `sys_platform == 'linux'` marker, since it needs GTK and won't
+> pip-install elsewhere).
 
 You need at least one backend extra (or none if you only plan to use
 cloud backends *and* already have the `openai` package). The `groq`
 backend reuses the `openai` client, so `[openai]` covers both.
 
+## Windows
+
+Windows 11 is tested and working on Python 3.14 (64-bit). Every
+dependency — `onnxruntime`, `faster-whisper`/`ctranslate2`,
+`pystray`/`Pillow`, `pynput` — resolves a ready-made `win_amd64` wheel,
+so there is no build toolchain to install and no need to downgrade
+Python.
+
+From PowerShell:
+
+```powershell
+py -m venv .venv
+.\.venv\Scripts\Activate.ps1
+# If activation is blocked by the execution policy, run once:
+#   Set-ExecutionPolicy -Scope CurrentUser RemoteSigned
+pip install -e .[whisper]      # or [all], or a cloud backend like [openai]
+scribe
+```
+
+That's the whole setup. There are **no system packages** to install
+(`apt`/`portaudio19-dev`/`xclip` are Linux-only) and **nothing to create
+by hand** — earlier builds needed a manual `C:\tmp` folder, which is no
+longer the case.
+
+- **Tray icon:** appears under the taskbar overflow arrow (`^`) by
+  default; pin it via *Settings → Personalization → Taskbar → Other
+  system tray icons*. A single click on the icon starts recording.
+- **Microphone:** if recording fails, enable *Settings → Privacy &
+  security → Microphone → "Let desktop apps access your microphone"*.
+
 ## Ubuntu / GNOME tray dependencies
 
 The tray icon needs system libraries for the AppIndicator stack:

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/docs/tray.md

@@ -26,9 +26,17 @@ recording / waiting, transcribing, and idle.
 Transcription and API errors are surfaced as a pop-up dialog instead
 of just crashing the tray.
 
-The tray requires `pystray` (and on Linux, `PyGObject` plus the
-appindicator system libs — see [installation.md](installation.md)).
-This is included with `pip install scribe-cli[all]` or `[app]`.
+The tray uses `pystray`, which is a **base dependency** — it ships with
+the plain `pip install scribe-cli`, so the default tray works on Windows
+and macOS with no extras. On Linux the AppIndicator backend additionally
+needs `PyGObject` plus the appindicator system libs; install those via
+`[app]` or `[all]` (see [installation.md](installation.md)).
+
+**Click behaviour differs by platform.** On Windows, a single click on
+the tray icon fires the default action (Record), because pystray's Win32
+backend can activate the default menu item. On Ubuntu the AppIndicator
+backend doesn't support a click-to-default action, so a click only opens
+the menu. This is a pystray backend difference, not a scribe bug.
 
 You can predefine which models appear in the tray menu with
 `--vosk-models`, `--whisper-models`, and `--whisper-futo-models`:
@@ -106,8 +114,46 @@ disabled "<vendor> — <reason>" row so you know what's missing.
 
 ## Global hotkey integration
 
-In tray / app mode scribe writes its PID to a pidfile and listens for
-two signals:
+### In-app global hotkeys
+
+In tray / app mode scribe runs a built-in global-hotkey listener (via
+`pynput`, a base dependency). Defaults are **OS-specific** so they land on
+chords that are normally free on each platform:
+
+| Action          | Linux (X11)   | Windows / macOS     |
+|-----------------|---------------|---------------------|
+| toggle record   | `Super`+`C`   | `Ctrl`+`Alt`+`C`    |
+| cancel          | `Super`+`Z`   | `Ctrl`+`Alt`+`Z`    |
+
+On Windows the Win-key chords are avoided on purpose: `Win`+`C` (Copilot)
+and `Win`+`Z` (Snap Layouts) are claimed by the shell, and pynput's hook
+doesn't suppress the key, so *both* the OS action and scribe's callback
+would fire. On macOS `Cmd` chords collide with copy/undo. The combos are
+**configurable** either way:
+
+```bash
+scribe --hotkey-record "<cmd>+c" --hotkey-cancel "<cmd>+z"   # Super/Win/Cmd + C/Z
+scribe --no-hotkeys                                          # turn the listener off
+```
+
+(`<cmd>` is the Super / Windows / Command key in pynput syntax.)
+
+**Platform support is uneven** — this is why scribe also keeps the Unix
+signal mechanism below:
+
+| Platform        | In-app global hotkeys                                                |
+|-----------------|---------------------------------------------------------------------|
+| Windows         | Works out of the box (default `Ctrl`+`Alt`+`C/Z`).                  |
+| Linux (X11)     | Works out of the box (default `Super`+`C/Z`).                       |
+| Linux (Wayland) | **Doesn't work** — the compositor blocks global key capture. Use the SIGUSR1/2 + custom-shortcut path below instead. |
+| macOS           | Needs Accessibility / Input-Monitoring permission (System Settings → Privacy & Security). |
+
+The listener is best-effort: if the OS won't grant a global hook, scribe
+logs a line and keeps running (tray + signals still work).
+
+### Unix signals (Linux / macOS)
+
+scribe writes its PID to a pidfile and listens for two signals:
 
 - `SIGUSR1` — toggle recording (same as clicking Record / Stop).
 - `SIGUSR2` — cancel an in-flight recording.

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/pyproject.toml

@@ -21,12 +21,17 @@ dependencies = [
     "unidecode",
     "termcolor",
     "platformdirs",
-    "desktop-ai-core>=0.2.0",
+    "desktop-ai-core>=0.3.1",      # >=0.3.1 carries the portable-tempdir pidfile fix (no more /tmp on Windows)
     # Runs the bundled silero VAD ONNX model (~2 MB shipped in scribe_data).
     # In base deps so silero is available out of the box — see scribe/audio.py.
     # `faster-whisper` already pulls it transitively, so installing with
     # [whisper] is free; standalone adds ~57 MB which is trivial for an STT tool.
     "onnxruntime",
+    # Default run uses the keyboard typer (needs pynput) and the system-tray app
+    # (needs pystray). Both are small, have wheels on all platforms, and back the
+    # README's headline zero-config experience, so they live in base deps.
+    "pynput",
+    "pystray",
 ]
 
 classifiers = [
@@ -82,17 +87,23 @@ keywords = [
 ]
 
 [project.optional-dependencies]
-keyboard = ["pynput"]
+keyboard = ["pynput"]                          # kept for back-compat; pynput is a base dep now
 whisper = ["faster-whisper"]
-whisper-futo = ["pywhispercpp"]
+whisper-futo = ["pywhispercpp"]                # CPU wheels exist but have known Windows DLL
+                                               # issues; keep strictly opt-in (not in [all])
 vosk = ["vosk"]
-app = ["pystray", "PyGObject"]
+# pystray is a base dep now; [app] only adds the Linux-only AppIndicator binding.
+# PyGObject needs GTK system libraries and does not pip-install on Windows/macOS,
+# so the PEP 508 marker skips it everywhere except Linux.
+app = ["PyGObject; sys_platform == 'linux'"]
 openai = ["openai>=2.37.0,<3", "soundfile"]
 groq = ["openai>=2.37.0,<3", "soundfile"]
 # [vad] is now a no-op alias kept for back-compat (`pip install scribe-cli[vad]`
 # was the documented install before onnxruntime moved into base deps).
 vad = []
-all = ["pynput", "faster-whisper", "pywhispercpp", "openai>=2.37.0,<3", "soundfile", "vosk", "pystray"]
+# pynput + pystray are base deps; pywhispercpp intentionally omitted due to
+# Windows build/DLL friction (install via scribe-cli[whisper-futo] if wanted).
+all = ["faster-whisper", "openai>=2.37.0,<3", "soundfile", "vosk", "PyGObject; sys_platform == 'linux'"]
 
 
 [tool.setuptools]

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/scribe/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.1.0'
-__version_tuple__ = version_tuple = (1, 1, 0)
+__version__ = version = '1.1.1'
+__version_tuple__ = version_tuple = (1, 1, 1)
 
-__commit_id__ = commit_id = 'g9b8b835fd'
+__commit_id__ = commit_id = 'g7a2dcc7f5'

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/scribe/app.py

@@ -2,6 +2,7 @@ import os
 from pathlib import Path
 import tomllib
 import signal
+import sys
 import argparse
 import platformdirs
 from scribe.audio import Microphone
@@ -599,6 +600,17 @@ def get_parser():
                        help="Whisper models offered in the tray menu.")
     group.add_argument("--whisper-futo-models", nargs="*", default=whisper_futo_models,
                        help="FUTO ACFT Whisper models offered in the tray menu.")
+    _hk_record, _hk_cancel = _default_hotkeys()
+    group.add_argument("--hotkey-record", default=_hk_record,
+                       help="Global hotkey to toggle recording in tray mode "
+                            "(pynput combo syntax, e.g. '<cmd>+c', '<ctrl>+<alt>+c'; "
+                            "<cmd> is the Super/Windows/Command key). "
+                            "Default: %(default)s (OS-specific).")
+    group.add_argument("--hotkey-cancel", default=_hk_cancel,
+                       help="Global hotkey to cancel an in-flight recording in tray mode. "
+                            "Default: %(default)s (OS-specific).")
+    group.add_argument("--no-hotkeys", action="store_false", dest="hotkeys",
+                       help="Disable the in-app global hotkey listener (tray mode).")
 
     return parser
 
@@ -743,6 +755,82 @@ def start_recording(micro, session, o, callback=None, **greetings):
 
 
 
+def _default_hotkeys():
+    """Return OS-appropriate (record, cancel) global-hotkey combos.
+
+    Super+C / Super+Z are typically free on Linux/X11, but poor defaults
+    elsewhere: on Windows the Win key combos collide with the shell (Win+C →
+    Copilot, Win+Z → Snap Layouts) and pynput's non-suppressing hook lets *both*
+    the OS action and our callback fire; on macOS <cmd> combos clash with
+    copy/undo. So default to Ctrl+Alt+C / Ctrl+Alt+Z off Linux, where those
+    chords are normally unbound. All values stay overridable via the CLI.
+    """
+    if sys.platform == "linux":
+        return "<cmd>+c", "<cmd>+z"
+    return "<ctrl>+<alt>+c", "<ctrl>+<alt>+z"
+
+
+def _start_global_hotkeys(icon, app_state):
+    """Start a pynput global-hotkey listener bound to the configured combos.
+
+    Mirrors the SIGUSR1/SIGUSR2 handlers (toggle record / cancel). Platform
+    support is uneven and intentionally best-effort — the Unix signal path
+    remains the supported route where global capture is restricted:
+
+      * Windows: works (Win32 hooks). Avoid OS-reserved Win+key combos.
+      * Linux/X11: works.
+      * Linux/Wayland: the compositor blocks global key grabs, so the listener
+        won't fire — use SIGUSR1/2 bound to a desktop shortcut instead.
+      * macOS: needs Accessibility/Input-Monitoring permission.
+
+    Runs as a daemon thread so it never blocks shutdown. Returns the listener
+    (or None when disabled/unavailable) and never raises — a failure here
+    must not take down the tray.
+    """
+    o = app_state.o
+    if not getattr(o, "hotkeys", True):
+        return None
+
+    _rec_default, _cancel_default = _default_hotkeys()
+    record_combo = getattr(o, "hotkey_record", _rec_default)
+    cancel_combo = getattr(o, "hotkey_cancel", _cancel_default)
+
+    try:
+        from pynput import keyboard
+    except Exception as exc:  # pragma: no cover - pynput is a base dep
+        print(colored(f"Global hotkeys disabled (pynput unavailable): {exc}", "light_red"))
+        return None
+
+    def _on_record():
+        app_state.cb_record(icon, None)
+
+    def _on_cancel():
+        if icon._session.busy:
+            app_state.cb_cancel(icon, None)
+
+    mapping = {}
+    if record_combo:
+        mapping[record_combo] = _on_record
+    if cancel_combo:
+        mapping[cancel_combo] = _on_cancel
+    if not mapping:
+        return None
+
+    try:
+        listener = keyboard.GlobalHotKeys(mapping)
+        listener.daemon = True
+        listener.start()
+    except Exception as exc:
+        print(colored(f"Global hotkeys unavailable ({exc}); "
+                      "use the tray menu or SIGUSR1/2.", "light_red"))
+        return None
+
+    shown = " / ".join(c for c in (record_combo, cancel_combo) if c)
+    print(f"Global hotkeys: {colored(shown, 'light_blue', attrs=['bold'])} "
+          "(record / cancel)")
+    return listener
+
+
 def create_app(micro, app_state):
     """Construct the system-tray pystray Icon from the unified menu spec.
 
@@ -831,6 +919,10 @@ def create_app(micro, app_state):
         register_signal_toggle(signal.SIGUSR2,
                                lambda: icon._session.busy and app_state.cb_cancel(icon, None))
 
+    # In-app global hotkeys (cross-platform, incl. Windows where SIGUSR* don't
+    # exist). Kept on the icon so the reference lives as long as the tray does.
+    icon._hotkey_listener = _start_global_hotkeys(icon, app_state)
+
     return icon
 
 

{scribe_cli-1.1.0 → scribe_cli-1.1.1}/scribe/models.py

@@ -307,6 +307,32 @@ class AbstractTranscriber(STTBackend):
                         f"Cut at silence after {elapsed:.2f}s "
                         f"(silent {sil_dur:.2f}s)"
                     )
+                # Long-pause escape hatch. When silence reaches the
+                # context-reset threshold and we still haven't committed,
+                # flush whatever is in the buffer as long as it clears
+                # the lower stream_chunk_min floor (the Whisper-
+                # hallucination protection). This catches short
+                # utterances stranded below stream_first_chunk_min: the
+                # user clearly stopped talking — don't lose what they
+                # said waiting for a bootstrap window that's never
+                # going to fill. Also clears the rolling context, since
+                # a pause this long is the same topic-shift signal that
+                # the speech-resumption path uses for its own reset.
+                reset_threshold = (self.stream_context_reset_silence
+                                   * silence_break)
+                if (not math.isinf(reset_threshold)
+                        and sil_dur >= reset_threshold
+                        and buffer_ms >= self.stream_chunk_min * 1000):
+                    if self._streaming_context:
+                        self.log(
+                            f"Clearing chunk context after {sil_dur:.2f}s pause"
+                        )
+                        self.clear_streaming_context()
+                    raise SilenceDetected(
+                        f"Cut at long silence after {elapsed:.2f}s "
+                        f"(silent {sil_dur:.2f}s, "
+                        f"below first-chunk floor)"
+                    )
             else:
                 # Auto and Max never silence-cut. Auto defers the cut
                 # decision to the force-cut below (which picks the best

scribe_cli-1.1.1/scribe/typers/base.py

@@ -0,0 +1,63 @@
+import re
+from typing import Callable, Protocol, Tuple, Type, runtime_checkable
+
+import unidecode
+
+# Split a string into maximal runs of ASCII plus each individual non-ASCII
+# character, preserving order: "café là" -> ["caf", "é", " l", "à", ""]-ish.
+_ASCII_RUN_OR_CHAR = re.compile(r"[\x00-\x7f]+|[^\x00-\x7f]")
+
+
+def type_ascii_safe(
+    emit: Callable[[str], None],
+    text: str,
+    errors: Tuple[Type[BaseException], ...],
+) -> None:
+    """Type ``text`` via ``emit`` while degrading untypable characters to ASCII
+    **without re-emitting already-typed text**.
+
+    Keystroke typers (wtype / eitype / pynput) emit left-to-right and abort
+    mid-string when the active xkb layout can't produce a character — but the
+    prefix is already out by then. The naive recovery (retry the whole string
+    transliterated to ASCII) re-types that prefix, so a chunk like
+    "le message dicté" lands as "le message dict" + "le message dicte" — the
+    duplicated-prefix bug.
+
+    Instead, emit maximal ASCII runs in one call each (always layout-typeable)
+    and, for each individual non-ASCII character, try it raw and fall back to
+    its ASCII transliteration on ``errors``. Unicode survives on layouts that
+    support it; the rest degrades per-character with no duplication. Failures
+    on ASCII content are genuine (compositor / daemon) and propagate.
+    """
+    for token in _ASCII_RUN_OR_CHAR.findall(text):
+        if token.isascii():
+            # Always layout-typeable; a failure here is genuine (compositor /
+            # daemon down) — let it propagate.
+            emit(token)
+        else:
+            try:
+                emit(token)
+            except errors:
+                try:
+                    emit(unidecode.unidecode(token))
+                except errors:
+                    # Even the transliteration is unrenderable — skip this
+                    # one char rather than abort the whole transcript.
+                    pass
+
+
+@runtime_checkable
+class Typer(Protocol):
+    name: str
+
+    def compatible(self) -> bool:
+        """True iff the host OS / session could *in principle* run this backend
+        (ignores setup). False means the backend is structurally impossible
+        here — e.g. ydotool on macOS, wtype on GNOME. Used by the menu to
+        hide incompatible rows entirely. Distinct from ``available()``, which
+        further requires binaries / daemons / sockets to be set up."""
+        ...
+
+    def available(self) -> bool: ...
+    def type(self, text: str) -> None: ...
+    def paste(self) -> None: ...