Update README.md

Remove old widget-based Python UI (replaced by QML), unused build
scripts, NL/variable/webfont variants, old shaders, PNG icons
(replaced by SVGs), and standalone test files. Add build.bat and
build.spec for the bootstrapper build system.

README.md

@@ -24,7 +24,7 @@
 
 ## 📡 The Transmission
 
-We are witnessing the **enshittification** of the digital world. What were once thriving social commons are being walled off, strip-mined for data, and degraded into rent-seeking silos. Your voice is no longer your own; it is a training set for a corporate oracle that charges you for the privilege of listening.
+We are witnessing the **enshittification** of the digital world. What were once vibrant social commons are being walled off, strip-mined for data, and degraded into rent-seeking silos. Your voice is no longer your own; it is a training set for a corporate oracle that charges you for the privilege of listening.
 
 **Whisper Voice** is a small act of sabotage against this trend.
 
@@ -59,7 +59,7 @@ The core engine (`CTranslate2`) is heavily optimized for Nvidia tensor cores.
 
 ## 🖋️ Universal Transcription
 
-Whisper Voice is the bridge between thought and text. It listens with superhuman precision, converting spoken word into written form across **99 languages**.
+At its core, Whisper Voice is the ultimate bridge between thought and text. It listens with superhuman precision, converting spoken word into written form across **99 languages**.
 
 *   **Punctuation Mastery**: Automatically handles capitalization and complex punctuation formatting.
 *   **Contextual Intelligence**: Smarter than standard dictation; it understands the flow of sentences to resolve homophones and technical jargon ($1.5k vs "fifteen hundred dollars").
@@ -113,7 +113,7 @@ The agent runs silently in the background, waiting for your signal.
 
 ### Injection Protocols
 *   **Clipboard Paste**: Standard text injection. Instant, reliable.
-*   **Simulate Typing**: Mimics physical keystrokes at high speed (6000 CPM). Bypasses anti-paste restrictions and "protected" windows.
+*   **Simulate Typing**: Mimics physical keystrokes at superhuman speed (6000 CPM). Bypasses anti-paste restrictions and "protected" windows.
 
 <br>
 
@@ -143,7 +143,7 @@ For users with limited GPU memory (e.g., 4GB cards) or those running heavy games
 
 ## ♿ Accessibility (WCAG 2.2 AAA)
 
-Whisper Voice is built to be usable by everyone. The entire interface has been engineered to meet **WCAG 2.2 AAA** - the highest tier of accessibility compliance. This is not a checkbox exercise; it is a structural commitment.
+Whisper Voice is built to be usable by everyone. The entire interface has been engineered to meet **WCAG 2.2 AAA** — the highest tier of accessibility compliance. This is not a checkbox exercise; it is a structural commitment.
 
 ### Color & Contrast
 Every design token is calibrated for **Enhanced Contrast** (WCAG 1.4.6, 7:1 minimum):
@@ -156,7 +156,7 @@ Every design token is calibrated for **Enhanced Contrast** (WCAG 1.4.6, 7:1 mini
 | `borderSubtle` | 3:1 | Non-text contrast for borders and separators |
 
 ### Keyboard Navigation
-Full keyboard access - no mouse required:
+Full keyboard access — no mouse required:
 
 *   **Tab / Shift+Tab**: Navigate between all interactive controls (sliders, switches, buttons, dropdowns, text fields).
 *   **Arrow Keys**: Navigate the Settings sidebar tabs.
@@ -166,7 +166,7 @@ Full keyboard access - no mouse required:
 ### Screen Reader Support
 Every component is annotated with semantic roles and descriptive names:
 
-*   Buttons, sliders, checkboxes, combo boxes, text fields - all declare their `Accessible.role` and `Accessible.name`.
+*   Buttons, sliders, checkboxes, combo boxes, text fields — all declare their `Accessible.role` and `Accessible.name`.
 *   Switches report "on" / "off" state in their accessible name.
 *   The loader status uses `AlertMessage` for live-region announcements.
 *   Settings tabs use `Tab` / `PageTab` roles matching WAI-ARIA patterns.
@@ -239,8 +239,8 @@ The engine understands the following 99 languages. You can lock the focus to a s
 ### ⚖️ PUBLIC DOMAIN (CC0 1.0)
 *No Rights Reserved. No Gods. No Masters. No Managers.*
 
+This project is developed with the help of a locally-run LLM via opencode for scaffolding, planning, and routine code tasks. Architecture and design decisions are my own.
+
 Credit to **OpenAI** (Whisper), **Systran** (Faster-Whisper), and **Silero** (VAD).
 
 </div>
-
-<sub><i>This project is developed with the help of a locally-run LLM via opencode for scaffolding, planning, and routine code tasks. Architecture and design decisions are my own.</i></sub>

RELEASE_NOTES.md

@@ -0,0 +1,28 @@
+# Release v1.0.4
+
+**"The Compatibility Update"**
+
+This release focuses on maximum stability across different hardware configurations (AMD, Intel, Nvidia) and fixing startup crashes related to corrupted models or missing drivers.
+
+## 🛠️ Critical Fixes
+
+### 1. Robust CPU Fallback (AMD / Intel Support)
+*   **Problem**: Previously, if an AMD user tried to run the app, it would crash instantly because it tried to load Nvidia CUDA libraries by default.
+*   **Fix**: The app now **silently detects** if CUDA initialization fails (due to missing DLLs or incompatible hardware) and **automatically falls back to CPU mode**.
+*   **Result**: The app "just works" on any Windows machine, regardless of GPU.
+
+### 2. Startup Crash Protection
+*   **Problem**: If `faster_whisper` was imported before checking for valid drivers, the app would crash on launch for some users.
+*   **Fix**: Implemented **Lazy Loading** for the AI engine. The app now starts the UI first, and only loads the heavy AI libraries inside a safety block that catches errors.
+
+### 3. Corrupt Model Auto-Repair
+*   **Problem**: Interrupted downloads could leave a corrupted model folder, preventing the app from ever starting again.
+*   **Fix**: If the app detects a "vocabulary missing" or invalid config error, it will now **automatically delete the corrupt folder** and allow you to re-download it cleanly.
+
+### 4. Windows DLL Injection
+*   **Fix**: Added explicit DLL path injection for `nvidia-cublas` and `nvidia-cudnn` to ensure Python 3.8+ can find the required CUDA libraries on Windows systems that don't have them in PATH.
+
+## 📦 Installation
+1.  Download `WhisperVoice.exe` below.
+2.  Replace your existing `.exe`.
+3.  Run it.

build.spec

@@ -1,5 +1,5 @@
 # -*- mode: python ; coding: utf-8 -*-
-# WhisperVoice - Single-file portable bootstrapper
+# WhisperVoice — Single-file portable bootstrapper
 #
 # This builds a TINY exe that contains only:
 #   - The bootstrapper (downloads Python + deps on first run)
@@ -8,7 +8,6 @@
 # NO heavy dependencies (torch, PySide6, etc.) are bundled.
 
 import os
-import sys
 import glob
 
 block_cipher = None
@@ -23,7 +22,7 @@ app_datas.append(('main.py', 'app_source'))
 # requirements.txt
 app_datas.append(('requirements.txt', 'app_source'))
 
-# src/**/*.py  (core, ui, utils - preserving directory structure)
+# src/**/*.py  (core, ui, utils — preserving directory structure)
 for py in glob.glob('src/**/*.py', recursive=True):
     dest = os.path.join('app_source', os.path.dirname(py))
     app_datas.append((py, dest))
@@ -43,7 +42,7 @@ for f in glob.glob(os.path.join(qml_dir, 'fonts', 'ttf', '*.ttf')):
 if os.path.exists(os.path.join('assets', 'icon.ico')):
     app_datas.append((os.path.join('assets', 'icon.ico'), os.path.join('app_source', 'assets')))
 
-# ── Analysis - only the bootstrapper, NO heavy imports ────────────────────
+# ── Analysis — only the bootstrapper, NO heavy imports ────────────────────
 
 a = Analysis(
     ['bootstrapper.py'],
@@ -55,14 +54,14 @@ a = Analysis(
     hooksconfig={},
     runtime_hooks=[],
     excludes=[
-        # Exclude everything heavy - the bootstrapper only uses stdlib
+        # Exclude everything heavy — the bootstrapper only uses stdlib
         'torch', 'numpy', 'scipy', 'PySide6', 'shiboken6',
         'faster_whisper', 'ctranslate2', 'llama_cpp',
         'sounddevice', 'soundfile', 'keyboard', 'pyperclip',
         'psutil', 'pynvml', 'pystray', 'PIL', 'Pillow',
         'darkdetect', 'huggingface_hub', 'requests',
         'tqdm', 'onnxruntime', 'av',
-        'matplotlib', 'notebook', 'IPython',
+        'tkinter', 'matplotlib', 'notebook', 'IPython',
     ],
     win_no_prefer_redirects=False,
     win_private_assemblies=False,
@@ -86,11 +85,11 @@ exe = EXE(
     bootloader_ignore_signals=False,
     strip=False,
     upx=True,
-    console=False,       # No console - bootstrapper allocates one when needed
+    console=False,       # No console — bootstrapper allocates one when needed
     disable_windowed_traceback=False,
     argv_emulation=False,
     target_arch=None,
     codesign_identity=None,
     entitlements_file=None,
-    icon='assets/icon.ico' if sys.platform == 'win32' else None,
+    icon='assets/icon.ico',
 )

main.py

@@ -9,14 +9,6 @@ app_dir = os.path.dirname(os.path.abspath(__file__))
 if app_dir not in sys.path:
     sys.path.insert(0, app_dir)
 
-# -----------------------------------------------------------------------------
-# LINUX: Force XWayland (X11) for reliable window positioning & overlay behavior.
-# Our input stack (evdev, UInput, wl-copy) is compositor-agnostic so this is safe.
-# Native Wayland lacks app-controlled window positioning which the overlay needs.
-# -----------------------------------------------------------------------------
-if sys.platform == 'linux' and os.environ.get('WAYLAND_DISPLAY'):
-    os.environ.setdefault('QT_QPA_PLATFORM', 'xcb')
-
 # -----------------------------------------------------------------------------
 # WINDOWS DLL FIX (CRITICAL for Portable CUDA)
 # Python 3.8+ on Windows requires explicit DLL directory addition.
@@ -24,32 +16,20 @@ if sys.platform == 'linux' and os.environ.get('WAYLAND_DISPLAY'):
 if os.name == 'nt' and hasattr(os, 'add_dll_directory'):
     try:
         from pathlib import Path
-        _candidate_dirs = set()
-
-        # 1. From sys.path (scan ALL site-packages, not just the first)
+        # Scan sys.path for site-packages
         for p in sys.path:
             path_obj = Path(p)
             if path_obj.name == 'site-packages' and path_obj.exists():
-                _candidate_dirs.add(str(path_obj.resolve()))
-
-        # 2. Relative to the Python executable (critical for embedded Python)
-        _exe_dir = Path(sys.executable).parent
-        for _sp in [_exe_dir / "Lib" / "site-packages", _exe_dir / "lib" / "site-packages"]:
-            if _sp.exists():
-                _candidate_dirs.add(str(_sp.resolve()))
-
-        # 3. Scan all candidates for nvidia DLL directories
-        for _sp_str in _candidate_dirs:
-            nvidia_path = Path(_sp_str) / "nvidia"
-            if nvidia_path.exists():
-                for subdir in nvidia_path.iterdir():
-                    bin_path = subdir / "bin"
-                    if bin_path.exists():
-                        os.add_dll_directory(str(bin_path))
-                        # Also add to PATH as fallback - some libraries
-                        # (e.g. CTranslate2) load DLLs lazily via LoadLibrary
-                        # which may not respect os.add_dll_directory()
-                        os.environ['PATH'] = str(bin_path) + os.pathsep + os.environ.get('PATH', '')
+                nvidia_path = path_obj / "nvidia"
+                if nvidia_path.exists():
+                    for subdir in nvidia_path.iterdir():
+                        # Add 'bin' folder from each nvidia stub (cublas, cudnn, etc.)
+                        bin_path = subdir / "bin"
+                        if bin_path.exists():
+                            os.add_dll_directory(str(bin_path))
+                # Also try adding site-packages itself just in case
+                # os.add_dll_directory(str(path_obj))
+                break
     except Exception:
         pass
 # -----------------------------------------------------------------------------
@@ -69,8 +49,7 @@ from src.core.hotkey_manager import HotkeyManager
 from src.core.config import ConfigManager
 from src.utils.injector import InputInjector
 from src.core.paths import get_models_path, get_bundle_path
-if os.name == 'nt':
-    from src.utils.window_hook import WindowHook
+from src.utils.window_hook import WindowHook
 
 from PySide6.QtGui import QSurfaceFormat
 
@@ -84,52 +63,37 @@ os.environ["QT_ENABLE_HIGHDPI_SCALING"] = "1"
 os.environ["QT_AUTOSCREENSCALEFACTOR"] = "1"
 
 # Detect resolution without creating QApplication (Fixes crash)
-if os.name == 'nt':
-    try:
-        import ctypes
-        user32 = ctypes.windll.user32
-        # Get physical screen width (unscaled)
-        # SetProcessDPIAware is needed to get the true resolution
-        user32.SetProcessDPIAware()
-        width = user32.GetSystemMetrics(0)
-        # Base scale centers around 1920 width.
-        # At 3840 (4k), res_scale is 2.0. If we want it ~40% smaller, we multiply by 0.6 = 1.2
-        res_scale = (width / 1920)
-        if width >= 3840:
-            res_scale *= 0.6 # Make it significantly smaller at 4k as requested
+try:
+    import ctypes
+    user32 = ctypes.windll.user32
+    # Get physical screen width (unscaled)
+    # SetProcessDPIAware is needed to get the true resolution
+    user32.SetProcessDPIAware()
+    width = user32.GetSystemMetrics(0)
+    # Base scale centers around 1920 width.
+    # At 3840 (4k), res_scale is 2.0. If we want it ~40% smaller, we multiply by 0.6 = 1.2
+    res_scale = (width / 1920)
+    if width >= 3840:
+        res_scale *= 0.6 # Make it significantly smaller at 4k as requested
+    
+    os.environ["QT_SCALE_FACTOR"] = str(max(1.0, res_scale))
+except:
+    pass
 
-        os.environ["QT_SCALE_FACTOR"] = str(max(1.0, res_scale))
-    except:
-        pass
-# On Linux, Qt handles DPI automatically via QT_ENABLE_HIGHDPI_SCALING
-
-# Detect "Reduce Motion" preference
-if os.name == 'nt':
-    try:
-        import ctypes
-        SPI_GETCLIENTAREAANIMATION = 0x1042
-        animation_enabled = ctypes.c_bool(True)
-        ctypes.windll.user32.SystemParametersInfoW(
-            SPI_GETCLIENTAREAANIMATION, 0,
-            ctypes.byref(animation_enabled), 0
-        )
-        if not animation_enabled.value:
-            ConfigManager().data["reduce_motion"] = True
-            ConfigManager().save()
-    except Exception:
-        pass
-elif sys.platform == 'linux':
-    try:
-        import subprocess as _sp
-        result = _sp.run(
-            ['gsettings', 'get', 'org.gnome.desktop.interface', 'enable-animations'],
-            capture_output=True, text=True, timeout=2
-        )
-        if result.stdout.strip() == 'false':
-            ConfigManager().data["reduce_motion"] = True
-            ConfigManager().save()
-    except Exception:
-        pass
+# Detect Windows "Reduce Motion" preference
+try:
+    import ctypes
+    SPI_GETCLIENTAREAANIMATION = 0x1042
+    animation_enabled = ctypes.c_bool(True)
+    ctypes.windll.user32.SystemParametersInfoW(
+        SPI_GETCLIENTAREAANIMATION, 0,
+        ctypes.byref(animation_enabled), 0
+    )
+    if not animation_enabled.value:
+        ConfigManager().data["reduce_motion"] = True
+        ConfigManager().save()
+except Exception:
+    pass
 
 # Configure Logging
 class QmlLoggingHandler(logging.Handler, QObject):
@@ -176,33 +140,23 @@ class DownloadWorker(QThread):
     def run(self):
         try:
             import requests
-            import shutil
+            from tqdm import tqdm
             model_path = get_models_path()
+            # Determine what to download
             dest_dir = model_path / f"faster-whisper-{self.model_name}"
             repo_id = f"Systran/faster-whisper-{self.model_name}"
-            required_files = ["config.json", "model.bin", "tokenizer.json", "vocabulary.json"]
+            files = ["config.json", "model.bin", "tokenizer.json", "vocabulary.json"]
             base_url = f"https://huggingface.co/{repo_id}/resolve/main"
 
-            # Skip if already complete
-            if dest_dir.exists() and all((dest_dir / f).exists() for f in required_files):
-                logging.info(f"Model {self.model_name} already downloaded.")
-                self.finished.emit()
-                return
-
-            # Download to a temp dir first, move on success
-            tmp_dir = model_path / f".tmp-faster-whisper-{self.model_name}"
-            if tmp_dir.exists():
-                shutil.rmtree(tmp_dir)
-            tmp_dir.mkdir(parents=True, exist_ok=True)
-
+            dest_dir.mkdir(parents=True, exist_ok=True)
             logging.info(f"Downloading {self.model_name} to {dest_dir}...")
-
+            
             # 1. Calculate Total Size
             total_size = 0
             file_sizes = {}
-
+            
             with requests.Session() as s:
-                for fname in required_files:
+                for fname in files:
                     url = f"{base_url}/{fname}"
                     head = s.head(url, allow_redirects=True)
                     if head.status_code == 200:
@@ -210,48 +164,42 @@ class DownloadWorker(QThread):
                         file_sizes[fname] = size
                         total_size += size
                     else:
-                        logging.warning(f"HEAD failed for {fname}: HTTP {head.status_code}")
-
-            # Abort if any required file is unavailable
-            missing = [f for f in required_files if f not in file_sizes]
-            if missing:
-                shutil.rmtree(tmp_dir, ignore_errors=True)
-                raise RuntimeError(f"Required model files unavailable: {missing}")
-
+                        # Fallback for vocabulary.json vs vocabulary.txt
+                        if fname == "vocabulary.json":
+                             # Try .txt? Or just skip if not found? 
+                             # Faster-whisper usually has vocabulary.json
+                             pass
+            
             # 2. Download loop
             downloaded_bytes = 0
-
+            
             with requests.Session() as s:
-                for fname in required_files:
+                for fname in files:
+                    if fname not in file_sizes: continue
+                    
                     url = f"{base_url}/{fname}"
-                    dest_file = tmp_dir / fname
-
+                    dest_file = dest_dir / fname
+                    
+                    # Resume check? 
+                    # Simpler to just overwrite for reliability unless we want complex resume logic.
+                    # We'll overwrite.
+                    
                     resp = s.get(url, stream=True)
                     resp.raise_for_status()
-
+                    
                     with open(dest_file, 'wb') as f:
                         for chunk in resp.iter_content(chunk_size=8192):
                             if chunk:
                                 f.write(chunk)
                                 downloaded_bytes += len(chunk)
+                                
+                                # Emit Progress
                                 if total_size > 0:
                                     pct = int((downloaded_bytes / total_size) * 100)
                                     self.progress.emit(pct)
 
-            # 3. Validate all files present and non-empty
-            for fname in required_files:
-                fpath = tmp_dir / fname
-                if not fpath.exists() or fpath.stat().st_size == 0:
-                    shutil.rmtree(tmp_dir, ignore_errors=True)
-                    raise RuntimeError(f"Download incomplete: {fname} missing or empty")
-
-            # 4. Atomic move: replace dest with completed download
-            if dest_dir.exists():
-                shutil.rmtree(dest_dir)
-            tmp_dir.rename(dest_dir)
-
             self.finished.emit()
-
+            
         except Exception as e:
             logging.error(f"Download failed: {e}")
             self.error.emit(str(e))
@@ -363,8 +311,7 @@ class WhisperApp(QObject):
         self.bridge.llmDownloadRequested.connect(self.on_llm_download_requested)
         
         self.engine.rootContext().setContextProperty("ui", self.bridge)
-        self.engine.rootContext().setContextProperty("isLinux", sys.platform == 'linux')
-
+        
         # 2. Tray setup
         self.tray = SystemTray()
         self.tray.quit_requested.connect(self.quit_app)
@@ -427,7 +374,7 @@ class WhisperApp(QObject):
         self.overlay_root.setProperty("color", "transparent")
 
         self.center_overlay()
-
+        
         # Preload Settings (Invisible)
         logging.info("Preloading Settings window...")
         self.open_settings()
@@ -435,27 +382,23 @@ class WhisperApp(QObject):
             self.settings_root.setVisible(False)
         
         # Install Low-Level Window Hook for Transparent Hit Test
-        if os.name == 'nt':
-            try:
-                from src.utils.window_hook import WindowHook
-                hwnd = self.overlay_root.winId()
-                # Initial scale from config
-                scale = float(self.config.get("ui_scale"))
-
-                # Current Overlay Dimensions
-                win_w = int(460 * scale)
-                win_h = int(180 * scale)
-
-                self.window_hook = WindowHook(hwnd, win_w, win_h, initial_scale=scale)
-                self.window_hook.install()
-
-                # Initial state: Disabled because we start inactive
-                self.window_hook.set_enabled(False)
-            except Exception as e:
-                logging.error(f"Failed to install WindowHook: {e}")
-        else:
-            # On Linux, use Qt flag for click-through overlay
-            self.overlay_root.setFlag(Qt.WindowTransparentForInput, True)
+        try:
+            from src.utils.window_hook import WindowHook
+            hwnd = self.overlay_root.winId()
+            # Initial scale from config
+            scale = float(self.config.get("ui_scale"))
+            
+            # Current Overlay Dimensions
+            win_w = int(460 * scale)
+            win_h = int(180 * scale)
+            
+            self.window_hook = WindowHook(hwnd, win_w, win_h, initial_scale=scale)
+            self.window_hook.install()
+            
+            # Initial state: Disabled because we start inactive
+            self.window_hook.set_enabled(False)
+        except Exception as e:
+            logging.error(f"Failed to install WindowHook: {e}")
 
     def center_overlay(self):
         """Calculates and sets the Overlay position above the taskbar."""
@@ -869,23 +812,19 @@ class WhisperApp(QObject):
         self.bridge.update_status("Error")
         logging.error(f"Download Error: {err}")
 
-    def _update_overlay_state(self, is_active):
-        """Update overlay visibility and input handling based on active state."""
-        if hasattr(self, 'window_hook'):
-            self.window_hook.set_enabled(is_active)
-        elif sys.platform == 'linux' and self.overlay_root:
-            self.overlay_root.setFlag(Qt.WindowTransparentForInput, not is_active)
-
     @Slot(bool)
     def on_ui_toggle_request(self, is_recording):
         """Called when recording state changes."""
+        # Update Window Hook to allow clicking if active
         is_active = is_recording or self.bridge.isProcessing
-        self._update_overlay_state(is_active)
-
+        if hasattr(self, 'window_hook'):
+            self.window_hook.set_enabled(is_active)
+            
     @Slot(bool)
     def on_processing_changed(self, is_processing):
         is_active = self.bridge.isRecording or is_processing
-        self._update_overlay_state(is_active)
+        if hasattr(self, 'window_hook'):
+            self.window_hook.set_enabled(is_active)
 
 if __name__ == "__main__":
     import sys

src/core/llm_engine.py

@@ -145,7 +145,7 @@ class LLMEngine:
                 logging.warning(f"LLM Refusal Detected: '{result}'. Falling back to original.")
                 return text # Return original text on refusal!
             
-            # --- Post-Processing ---
+            # --- Robust Post-Processing ---
             
             # 1. Strip quotes 
             if result.startswith('"') and result.endswith('"') and len(result) > 2 and '"' not in result[1:-1]: