supervertaler 1.9.163__py3-none-any.whl

modules/voice_commands.py

@@ -0,0 +1,920 @@
+"""
+Voice Commands Module for Supervertaler
+Talon-style voice command system with 3 tiers:
+- Tier 1: In-app commands (Python/PyQt6)
+- Tier 2: System commands (AutoHotkey scripts)
+- Tier 3: Dictation fallback (insert as text)
+"""
+
+import json
+import os
+import re
+import subprocess
+from pathlib import Path
+from typing import Optional, Dict, List, Callable, Tuple
+from dataclasses import dataclass, field
+from difflib import SequenceMatcher
+from PyQt6.QtCore import QObject, pyqtSignal
+
+
+@dataclass
+class VoiceCommand:
+    """Represents a single voice command"""
+    phrase: str  # The spoken phrase (e.g., "confirm segment")
+    aliases: List[str] = field(default_factory=list)  # Alternative phrases
+    action_type: str = "internal"  # "internal", "keystroke", "ahk_script", "ahk_inline"
+    action: str = ""  # Action to execute
+    description: str = ""  # Human-readable description
+    category: str = "general"  # Category for organization
+    enabled: bool = True
+    
+    def matches(self, spoken_text: str, threshold: float = 0.85) -> Tuple[bool, float]:
+        """
+        Check if spoken text matches this command.
+        Returns (is_match, confidence_score)
+        """
+        spoken_lower = spoken_text.lower().strip()
+        
+        # Check exact matches first
+        all_phrases = [self.phrase.lower()] + [a.lower() for a in self.aliases]
+        for phrase in all_phrases:
+            if spoken_lower == phrase:
+                return (True, 1.0)
+        
+        # Check fuzzy matches
+        best_score = 0.0
+        for phrase in all_phrases:
+            # Use SequenceMatcher for fuzzy matching
+            score = SequenceMatcher(None, spoken_lower, phrase).ratio()
+            best_score = max(best_score, score)
+            
+            # Also check if spoken text contains the phrase
+            if phrase in spoken_lower or spoken_lower in phrase:
+                # Boost score for partial matches
+                length_ratio = min(len(phrase), len(spoken_lower)) / max(len(phrase), len(spoken_lower))
+                best_score = max(best_score, 0.9 * length_ratio)
+        
+        return (best_score >= threshold, best_score)
+
+
+class VoiceCommandManager(QObject):
+    """
+    Manages voice commands - matching spoken text to actions and executing them.
+    """
+    
+    # Signals
+    command_executed = pyqtSignal(str, str)  # (command_phrase, result_message)
+    command_not_found = pyqtSignal(str)  # spoken_text that didn't match
+    error_occurred = pyqtSignal(str)  # error message
+    
+    # Default commands
+    DEFAULT_COMMANDS = [
+        # Navigation
+        VoiceCommand("next segment", ["next", "down"], "internal", "navigate_next", 
+                    "Move to next segment", "navigation"),
+        VoiceCommand("previous segment", ["previous", "back", "up"], "internal", "navigate_previous",
+                    "Move to previous segment", "navigation"),
+        VoiceCommand("first segment", ["go to start", "beginning"], "internal", "navigate_first",
+                    "Jump to first segment", "navigation"),
+        VoiceCommand("last segment", ["go to end", "end"], "internal", "navigate_last",
+                    "Jump to last segment", "navigation"),
+        
+        # Segment actions
+        VoiceCommand("confirm", ["confirm segment", "done", "okay"], "internal", "confirm_segment",
+                    "Confirm current segment", "editing"),
+        VoiceCommand("copy source", ["copy from source", "source to target"], "internal", "copy_source_to_target",
+                    "Copy source text to target", "editing"),
+        VoiceCommand("clear target", ["clear", "delete target"], "internal", "clear_target",
+                    "Clear target text", "editing"),
+        VoiceCommand("undo", [], "keystroke", "ctrl+z",
+                    "Undo last action", "editing"),
+        VoiceCommand("redo", [], "keystroke", "ctrl+y",
+                    "Redo last action", "editing"),
+        
+        # Translation
+        VoiceCommand("translate", ["translate segment", "translate this"], "internal", "translate_segment",
+                    "AI translate current segment", "translation"),
+        VoiceCommand("translate all", ["batch translate"], "internal", "batch_translate",
+                    "Translate all segments", "translation"),
+        
+        # Lookup & Search
+        VoiceCommand("lookup", ["super lookup", "search"], "internal", "open_superlookup",
+                    "Open Superlookup (Ctrl+K)", "lookup"),
+        VoiceCommand("concordance", ["search memory", "search TM"], "internal", "concordance_search",
+                    "Open concordance search", "lookup"),
+        
+        # File operations
+        VoiceCommand("save project", ["save"], "keystroke", "ctrl+s",
+                    "Save current project", "file"),
+        VoiceCommand("open project", ["open"], "keystroke", "ctrl+o",
+                    "Open project", "file"),
+        
+        # View
+        VoiceCommand("show log", ["open log", "log tab"], "internal", "show_log",
+                    "Show log panel", "view"),
+        VoiceCommand("show editor", ["editor tab", "go to editor"], "internal", "show_editor",
+                    "Show editor panel", "view"),
+        
+        # Dictation control
+        VoiceCommand("start dictation", ["dictate", "voice input"], "internal", "start_dictation",
+                    "Start voice dictation mode", "dictation"),
+        VoiceCommand("stop listening", ["stop", "pause"], "internal", "stop_listening",
+                    "Stop voice recognition", "dictation"),
+        
+        # memoQ-specific (AHK)
+        VoiceCommand("glossary", ["add term", "add to glossary"], "ahk_inline", 
+                    "Send, !{Down}",  # Alt+Down
+                    "Add term pair to memoQ termbase", "memoq"),
+        VoiceCommand("tag next", ["next tag", "insert tag"], "ahk_inline",
+                    "Send, ^{PgDn}\nSleep, 100\nSend, {F9}\nSleep, 100\nSend, ^{Enter}",
+                    "Go to end, insert next tag, confirm", "memoq"),
+        VoiceCommand("confirm memoQ", ["confirm memo"], "ahk_inline",
+                    "Send, ^{Enter}",
+                    "Confirm segment in memoQ", "memoq"),
+        
+        # Trados-specific (AHK)
+        VoiceCommand("confirm trados", ["confirm studio"], "ahk_inline",
+                    "Send, ^{Enter}",
+                    "Confirm segment in Trados Studio", "trados"),
+    ]
+    
+    def __init__(self, user_data_path: Path, main_window=None):
+        super().__init__()
+        self.user_data_path = user_data_path
+        self.main_window = main_window
+        self.commands: List[VoiceCommand] = []
+        self.commands_file = user_data_path / "voice_commands.json"
+        self.ahk_script_dir = user_data_path / "voice_scripts"
+        self.match_threshold = 0.85  # Minimum similarity for fuzzy matching
+        
+        # Internal action handlers (mapped to main_window methods)
+        self.internal_handlers: Dict[str, Callable] = {}
+        
+        # Ensure directories exist
+        self.ahk_script_dir.mkdir(parents=True, exist_ok=True)
+        
+        # Load commands
+        self.load_commands()
+        
+        # Register internal handlers if main_window provided
+        if main_window:
+            self.register_main_window_handlers(main_window)
+    
+    def register_main_window_handlers(self, main_window):
+        """Register handlers that call main window methods"""
+        self.main_window = main_window
+        
+        self.internal_handlers = {
+            # Navigation - using correct method names from Supervertaler.py
+            "navigate_next": lambda: main_window.go_to_next_segment() if hasattr(main_window, 'go_to_next_segment') else self._log_missing('go_to_next_segment'),
+            "navigate_previous": lambda: main_window.go_to_previous_segment() if hasattr(main_window, 'go_to_previous_segment') else self._log_missing('go_to_previous_segment'),
+            "navigate_first": lambda: main_window.go_to_first_segment() if hasattr(main_window, 'go_to_first_segment') else self._log_missing('go_to_first_segment'),
+            "navigate_last": lambda: main_window.go_to_last_segment() if hasattr(main_window, 'go_to_last_segment') else self._log_missing('go_to_last_segment'),
+            
+            # Editing - confirm_and_next_unconfirmed is the Enter key behavior
+            "confirm_segment": lambda: main_window.confirm_and_next_unconfirmed() if hasattr(main_window, 'confirm_and_next_unconfirmed') else self._log_missing('confirm_and_next_unconfirmed'),
+            "copy_source_to_target": lambda: main_window.copy_source_to_grid_target() if hasattr(main_window, 'copy_source_to_grid_target') else self._log_missing('copy_source_to_grid_target'),
+            "clear_target": lambda: main_window.clear_grid_target() if hasattr(main_window, 'clear_grid_target') else self._log_missing('clear_grid_target'),
+            
+            # Translation
+            "translate_segment": lambda: main_window.translate_current_segment() if hasattr(main_window, 'translate_current_segment') else self._log_missing('translate_current_segment'),
+            "batch_translate": lambda: main_window.translate_batch() if hasattr(main_window, 'translate_batch') else self._log_missing('translate_batch'),
+            
+            # Lookup
+            "open_superlookup": lambda: main_window._go_to_superlookup() if hasattr(main_window, '_go_to_superlookup') else self._log_missing('_go_to_superlookup'),
+            "concordance_search": lambda: main_window.show_concordance_search() if hasattr(main_window, 'show_concordance_search') else self._log_missing('show_concordance_search'),
+            
+            # View
+            "show_log": lambda: self._show_tab(main_window, "Log"),
+            "show_editor": lambda: self._show_tab(main_window, "Editor"),
+            
+            # Dictation
+            "start_dictation": lambda: main_window.start_voice_dictation() if hasattr(main_window, 'start_voice_dictation') else self._log_missing('start_voice_dictation'),
+            "stop_listening": lambda: self._stop_voice_recognition(),
+        }
+    
+    def _log_missing(self, method_name: str):
+        """Log when a method is missing from main_window"""
+        print(f"⚠️ Voice command: Method '{method_name}' not found on main window")
+        if self.main_window and hasattr(self.main_window, 'log'):
+            self.main_window.log(f"⚠️ Voice command: Method '{method_name}' not found")
+    
+    def _show_tab(self, main_window, tab_name: str):
+        """Helper to switch to a specific tab"""
+        if hasattr(main_window, 'main_tabs'):
+            for i in range(main_window.main_tabs.count()):
+                if tab_name.lower() in main_window.main_tabs.tabText(i).lower():
+                    main_window.main_tabs.setCurrentIndex(i)
+                    return
+    
+    def _stop_voice_recognition(self):
+        """Stop the voice recognition system"""
+        if self.main_window and hasattr(self.main_window, 'voice_command_listener'):
+            listener = self.main_window.voice_command_listener
+            if listener and hasattr(listener, 'stop'):
+                listener.stop()
+    
+    def load_commands(self):
+        """Load commands from JSON file, or create defaults"""
+        if self.commands_file.exists():
+            try:
+                with open(self.commands_file, 'r', encoding='utf-8') as f:
+                    data = json.load(f)
+                
+                self.commands = []
+                self.match_threshold = data.get('match_threshold', 0.85)
+                
+                for cmd_data in data.get('commands', []):
+                    self.commands.append(VoiceCommand(
+                        phrase=cmd_data['phrase'],
+                        aliases=cmd_data.get('aliases', []),
+                        action_type=cmd_data.get('action_type', 'internal'),
+                        action=cmd_data.get('action', ''),
+                        description=cmd_data.get('description', ''),
+                        category=cmd_data.get('category', 'general'),
+                        enabled=cmd_data.get('enabled', True)
+                    ))
+                
+                return
+            except Exception as e:
+                print(f"Error loading voice commands: {e}")
+        
+        # Use defaults
+        self.commands = self.DEFAULT_COMMANDS.copy()
+        self.save_commands()
+    
+    def save_commands(self):
+        """Save commands to JSON file"""
+        data = {
+            'version': '1.0',
+            'match_threshold': self.match_threshold,
+            'commands': [
+                {
+                    'phrase': cmd.phrase,
+                    'aliases': cmd.aliases,
+                    'action_type': cmd.action_type,
+                    'action': cmd.action,
+                    'description': cmd.description,
+                    'category': cmd.category,
+                    'enabled': cmd.enabled
+                }
+                for cmd in self.commands
+            ]
+        }
+        
+        try:
+            with open(self.commands_file, 'w', encoding='utf-8') as f:
+                json.dump(data, f, indent=2, ensure_ascii=False)
+        except Exception as e:
+            self.error_occurred.emit(f"Failed to save voice commands: {e}")
+    
+    def find_matching_command(self, spoken_text: str) -> Optional[Tuple[VoiceCommand, float]]:
+        """
+        Find the best matching command for spoken text.
+        Returns (command, confidence) or None if no match.
+        """
+        spoken_text = spoken_text.strip()
+        if not spoken_text:
+            return None
+        
+        best_match = None
+        best_score = 0.0
+        
+        for cmd in self.commands:
+            if not cmd.enabled:
+                continue
+            
+            is_match, score = cmd.matches(spoken_text, self.match_threshold)
+            if is_match and score > best_score:
+                best_match = cmd
+                best_score = score
+        
+        if best_match:
+            return (best_match, best_score)
+        return None
+    
+    def execute_command(self, command: VoiceCommand) -> bool:
+        """Execute a voice command. Returns True on success."""
+        try:
+            if command.action_type == "internal":
+                return self._execute_internal(command)
+            elif command.action_type == "keystroke":
+                return self._execute_keystroke(command)
+            elif command.action_type == "ahk_script":
+                return self._execute_ahk_script(command)
+            elif command.action_type == "ahk_inline":
+                return self._execute_ahk_inline(command)
+            else:
+                self.error_occurred.emit(f"Unknown action type: {command.action_type}")
+                return False
+        except Exception as e:
+            import traceback
+            self.error_occurred.emit(f"Error executing '{command.phrase}': {e}\n{traceback.format_exc()}")
+            return False
+    
+    def _execute_internal(self, command: VoiceCommand) -> bool:
+        """Execute an internal Python action"""
+        handler = self.internal_handlers.get(command.action)
+        if handler:
+            try:
+                result = handler()
+                # Log success to main window if available
+                if self.main_window and hasattr(self.main_window, 'log'):
+                    self.main_window.log(f"✓ Voice command executed: {command.phrase} → {command.action}")
+                self.command_executed.emit(command.phrase, f"✓ {command.description}")
+                return True
+            except Exception as e:
+                import traceback
+                error_msg = f"Error in handler for '{command.phrase}': {e}"
+                if self.main_window and hasattr(self.main_window, 'log'):
+                    self.main_window.log(f"❌ {error_msg}")
+                    self.main_window.log(traceback.format_exc())
+                self.error_occurred.emit(error_msg)
+                return False
+        else:
+            error_msg = f"No handler for internal action: {command.action}"
+            if self.main_window and hasattr(self.main_window, 'log'):
+                self.main_window.log(f"❌ {error_msg}")
+                self.main_window.log(f"   Available handlers: {list(self.internal_handlers.keys())}")
+            self.error_occurred.emit(error_msg)
+            return False
+    
+    def _execute_keystroke(self, command: VoiceCommand) -> bool:
+        """Execute a keystroke via AutoHotkey"""
+        # Convert keystroke format (e.g., "ctrl+s") to AHK format
+        ahk_keys = self._convert_to_ahk_keys(command.action)
+        ahk_code = f"Send, {ahk_keys}"
+        return self._run_ahk_code(ahk_code, command)
+    
+    def _execute_ahk_script(self, command: VoiceCommand) -> bool:
+        """Execute a saved AHK script file"""
+        script_path = self.ahk_script_dir / f"{command.action}.ahk"
+        if not script_path.exists():
+            self.error_occurred.emit(f"AHK script not found: {script_path}")
+            return False
+        
+        try:
+            # Find AutoHotkey executable
+            ahk_exe = self._find_ahk_executable()
+            if not ahk_exe:
+                self.error_occurred.emit("AutoHotkey not found. Please install AutoHotkey v2.")
+                return False
+            
+            subprocess.Popen([ahk_exe, str(script_path)], 
+                           creationflags=subprocess.CREATE_NO_WINDOW)
+            self.command_executed.emit(command.phrase, f"✓ {command.description}")
+            return True
+        except Exception as e:
+            self.error_occurred.emit(f"Failed to run AHK script: {e}")
+            return False
+    
+    def _execute_ahk_inline(self, command: VoiceCommand) -> bool:
+        """Execute inline AHK code"""
+        return self._run_ahk_code(command.action, command)
+    
+    def _run_ahk_code(self, ahk_code: str, command: VoiceCommand) -> bool:
+        """Run arbitrary AHK code"""
+        try:
+            ahk_exe = self._find_ahk_executable()
+            if not ahk_exe:
+                self.error_occurred.emit("AutoHotkey not found. Please install AutoHotkey v2.")
+                return False
+            
+            # Create temporary script
+            temp_script = self.ahk_script_dir / "_temp_voice_cmd.ahk"
+            
+            # Wrap code in AHK v2 format
+            full_script = f"""#Requires AutoHotkey v2.0
+#SingleInstance Force
+{ahk_code}
+ExitApp
+"""
+            
+            with open(temp_script, 'w', encoding='utf-8') as f:
+                f.write(full_script)
+            
+            # Run script
+            subprocess.Popen([ahk_exe, str(temp_script)],
+                           creationflags=subprocess.CREATE_NO_WINDOW)
+            
+            self.command_executed.emit(command.phrase, f"✓ {command.description}")
+            return True
+            
+        except Exception as e:
+            self.error_occurred.emit(f"Failed to run AHK code: {e}")
+            return False
+    
+    def _convert_to_ahk_keys(self, keystroke: str) -> str:
+        """Convert keystroke string to AHK Send format"""
+        # Map modifier names to AHK symbols
+        modifiers = {
+            'ctrl': '^',
+            'control': '^',
+            'alt': '!',
+            'shift': '+',
+            'win': '#',
+            'windows': '#'
+        }
+        
+        # Special key names
+        special_keys = {
+            'enter': '{Enter}',
+            'return': '{Enter}',
+            'tab': '{Tab}',
+            'escape': '{Esc}',
+            'esc': '{Esc}',
+            'space': '{Space}',
+            'backspace': '{Backspace}',
+            'delete': '{Delete}',
+            'del': '{Delete}',
+            'insert': '{Insert}',
+            'ins': '{Insert}',
+            'home': '{Home}',
+            'end': '{End}',
+            'pageup': '{PgUp}',
+            'pgup': '{PgUp}',
+            'pagedown': '{PgDn}',
+            'pgdn': '{PgDn}',
+            'up': '{Up}',
+            'down': '{Down}',
+            'left': '{Left}',
+            'right': '{Right}',
+            'f1': '{F1}', 'f2': '{F2}', 'f3': '{F3}', 'f4': '{F4}',
+            'f5': '{F5}', 'f6': '{F6}', 'f7': '{F7}', 'f8': '{F8}',
+            'f9': '{F9}', 'f10': '{F10}', 'f11': '{F11}', 'f12': '{F12}',
+        }
+        
+        parts = keystroke.lower().replace(' ', '').split('+')
+        result = ''
+        
+        for part in parts:
+            if part in modifiers:
+                result += modifiers[part]
+            elif part in special_keys:
+                result += special_keys[part]
+            else:
+                # Regular key
+                result += part
+        
+        return result
+    
+    def _find_ahk_executable(self) -> Optional[str]:
+        """Find AutoHotkey v2 executable"""
+        # Common installation paths
+        possible_paths = [
+            r"C:\Program Files\AutoHotkey\v2\AutoHotkey64.exe",
+            r"C:\Program Files\AutoHotkey\v2\AutoHotkey32.exe",
+            r"C:\Program Files\AutoHotkey\AutoHotkey.exe",
+            r"C:\Program Files (x86)\AutoHotkey\AutoHotkey.exe",
+        ]
+        
+        # Check PATH first
+        import shutil
+        ahk_in_path = shutil.which("AutoHotkey64") or shutil.which("AutoHotkey")
+        if ahk_in_path:
+            return ahk_in_path
+        
+        # Check common locations
+        for path in possible_paths:
+            if os.path.exists(path):
+                return path
+        
+        return None
+    
+    def process_spoken_text(self, spoken_text: str) -> Tuple[bool, str]:
+        """
+        Process spoken text - try to match command, return success status and message.
+        Returns (was_command, message_or_text)
+        - If command matched: (True, "Command executed: ...")
+        - If no match: (False, original_spoken_text) for dictation fallback
+        """
+        match_result = self.find_matching_command(spoken_text)
+        
+        if match_result:
+            command, confidence = match_result
+            success = self.execute_command(command)
+            if success:
+                return (True, f"✓ {command.phrase} ({confidence:.0%})")
+            else:
+                return (True, f"✗ Failed: {command.phrase}")
+        
+        # No command matched - return text for dictation
+        self.command_not_found.emit(spoken_text)
+        return (False, spoken_text)
+    
+    def add_command(self, command: VoiceCommand):
+        """Add a new command"""
+        self.commands.append(command)
+        self.save_commands()
+    
+    def remove_command(self, phrase: str):
+        """Remove a command by phrase"""
+        self.commands = [c for c in self.commands if c.phrase != phrase]
+        self.save_commands()
+    
+    def get_commands_by_category(self) -> Dict[str, List[VoiceCommand]]:
+        """Get commands organized by category"""
+        categories: Dict[str, List[VoiceCommand]] = {}
+        for cmd in self.commands:
+            if cmd.category not in categories:
+                categories[cmd.category] = []
+            categories[cmd.category].append(cmd)
+        return categories
+    
+    def export_commands(self, filepath: Path):
+        """Export commands to a file"""
+        data = {
+            'version': '1.0',
+            'match_threshold': self.match_threshold,
+            'commands': [
+                {
+                    'phrase': cmd.phrase,
+                    'aliases': cmd.aliases,
+                    'action_type': cmd.action_type,
+                    'action': cmd.action,
+                    'description': cmd.description,
+                    'category': cmd.category,
+                    'enabled': cmd.enabled
+                }
+                for cmd in self.commands
+            ]
+        }
+        with open(filepath, 'w', encoding='utf-8') as f:
+            json.dump(data, f, indent=2, ensure_ascii=False)
+    
+    def import_commands(self, filepath: Path, merge: bool = True):
+        """Import commands from a file"""
+        with open(filepath, 'r', encoding='utf-8') as f:
+            data = json.load(f)
+        
+        imported_commands = []
+        for cmd_data in data.get('commands', []):
+            imported_commands.append(VoiceCommand(
+                phrase=cmd_data['phrase'],
+                aliases=cmd_data.get('aliases', []),
+                action_type=cmd_data.get('action_type', 'internal'),
+                action=cmd_data.get('action', ''),
+                description=cmd_data.get('description', ''),
+                category=cmd_data.get('category', 'general'),
+                enabled=cmd_data.get('enabled', True)
+            ))
+        
+        if merge:
+            # Add imported commands, skip duplicates
+            existing_phrases = {c.phrase for c in self.commands}
+            for cmd in imported_commands:
+                if cmd.phrase not in existing_phrases:
+                    self.commands.append(cmd)
+        else:
+            # Replace all commands
+            self.commands = imported_commands
+        
+        self.save_commands()
+
+
+class ContinuousVoiceListener(QObject):
+    """
+    Continuous voice listening with Voice Activity Detection (VAD).
+    
+    How it works:
+    1. Continuously monitors microphone audio levels
+    2. When speech is detected (audio above threshold), starts recording
+    3. When silence is detected (audio below threshold for X ms), stops recording
+    4. Sends recording to Whisper for transcription
+    5. Processes result (command or dictation)
+    6. Repeats
+    
+    This eliminates the need to press F9 twice - just speak and it listens.
+    """
+    
+    # Signals
+    listening_started = pyqtSignal()
+    listening_stopped = pyqtSignal()
+    speech_detected = pyqtSignal(str)  # Raw transcribed text
+    command_detected = pyqtSignal(str, str)  # (phrase, result)
+    text_for_dictation = pyqtSignal(str)  # Text that didn't match any command
+    status_update = pyqtSignal(str)
+    error_occurred = pyqtSignal(str)
+    vad_status_changed = pyqtSignal(str)  # "listening", "recording", "processing"
+    
+    def __init__(self, command_manager: VoiceCommandManager, 
+                 model_name: str = "base",
+                 language: str = "auto",
+                 use_api: bool = False,
+                 api_key: str = None):
+        super().__init__()
+        self.command_manager = command_manager
+        self.model_name = model_name
+        self.language = None if language == "auto" else language
+        self.use_api = use_api
+        self.api_key = api_key
+        
+        # VAD settings
+        self.speech_threshold = 0.02  # RMS threshold to detect speech (adjustable)
+        self.silence_duration = 0.8  # Seconds of silence before stopping recording
+        self.min_speech_duration = 0.3  # Minimum speech duration to process
+        self.max_speech_duration = 15.0  # Maximum recording duration
+        self.is_listening = False
+        self._thread = None
+        self._whisper_model = None  # Cached Whisper model
+        
+    def start(self):
+        """Start continuous listening"""
+        if self.is_listening:
+            return
+        
+        self.is_listening = True
+        self._thread = _VADListenerThread(self)
+        self._thread.transcription_ready.connect(self._on_transcription)
+        self._thread.status_update.connect(self.status_update.emit)
+        self._thread.error_occurred.connect(self.error_occurred.emit)
+        self._thread.vad_status.connect(self.vad_status_changed.emit)
+        self._thread.start()
+        self.listening_started.emit()
+    
+    def stop(self):
+        """Stop continuous listening"""
+        self.is_listening = False
+        if self._thread:
+            self._thread.stop()
+            self._thread = None
+        self.listening_stopped.emit()
+    
+    def set_sensitivity(self, level: str):
+        """
+        Set microphone sensitivity level.
+        - "low": Requires loud speech (noisy environment)
+        - "medium": Normal sensitivity
+        - "high": Picks up quiet speech (quiet environment)
+        """
+        thresholds = {
+            "low": 0.04,
+            "medium": 0.02,
+            "high": 0.01
+        }
+        self.speech_threshold = thresholds.get(level, 0.02)
+    
+    def _on_transcription(self, text: str):
+        """Handle transcribed speech"""
+        self.speech_detected.emit(text)
+        
+        # Try to match as command
+        was_command, result = self.command_manager.process_spoken_text(text)
+        
+        if was_command:
+            self.command_detected.emit(text, result)
+        else:
+            # Pass to dictation
+            self.text_for_dictation.emit(text)
+
+
+class _VADListenerThread(QObject):
+    """
+    Voice Activity Detection listener thread.
+    Uses amplitude-based VAD to detect speech start/end.
+    """
+    
+    transcription_ready = pyqtSignal(str)
+    status_update = pyqtSignal(str)
+    error_occurred = pyqtSignal(str)
+    vad_status = pyqtSignal(str)  # "waiting", "recording", "processing"
+    
+    def __init__(self, listener: ContinuousVoiceListener):
+        super().__init__()
+        self.listener = listener
+        self._running = False
+        self._thread = None
+        self._model = None  # Cached whisper model
+    
+    def start(self):
+        """Start the listener thread"""
+        import threading
+        
+        self._running = True
+        self._thread = threading.Thread(target=self._run, daemon=True)
+        self._thread.start()
+    
+    def stop(self):
+        """Stop the listener thread"""
+        self._running = False
+    
+    def _run(self):
+        """Main VAD listening loop"""
+        try:
+            import sounddevice as sd
+            import numpy as np
+            import tempfile
+            import wave
+            import os
+            import time
+            
+            # Sample rate and chunk settings
+            sample_rate = 16000
+            chunk_samples = int(0.1 * sample_rate)  # 100ms chunks for VAD
+            
+            # Get settings from listener
+            speech_threshold = self.listener.speech_threshold
+            silence_duration = self.listener.silence_duration
+            min_speech_duration = self.listener.min_speech_duration
+            max_speech_duration = self.listener.max_speech_duration
+            
+            # Check if using API or local model
+            if self.listener.use_api and self.listener.api_key:
+                self.status_update.emit("🎤 Using OpenAI Whisper API (fast & accurate)")
+                self._model = None  # No local model needed
+            else:
+                # Load local Whisper model once
+                self.status_update.emit("🎤 Loading local speech model...")
+                self.vad_status.emit("loading")
+                try:
+                    import whisper
+                except ImportError:
+                    self.error_occurred.emit(
+                        "Local Whisper is not installed.\n\n"
+                        "Option A (recommended): Choose 'OpenAI Whisper API' in Settings → Supervoice (requires OpenAI API key).\n"
+                        "Option B: Install Local Whisper:\n"
+                        "  pip install supervertaler[local-whisper]"
+                    )
+                    self._running = False
+                    return
+                self._model = whisper.load_model(self.listener.model_name)
+            
+            self.status_update.emit("🎤 Always-on listening active (waiting for speech...)")
+            self.vad_status.emit("waiting")
+            
+            # Audio buffer for recording
+            audio_buffer = []
+            is_recording = False
+            silence_start = None
+            speech_start = None
+            
+            def audio_callback(indata, frames, time_info, status):
+                """Callback for audio stream - processes each chunk"""
+                nonlocal audio_buffer, is_recording, silence_start, speech_start
+                
+                if not self._running:
+                    return
+                
+                # Calculate RMS amplitude
+                rms = np.sqrt(np.mean(indata**2))
+                is_speech = rms > speech_threshold
+                
+                if is_speech:
+                    if not is_recording:
+                        # Speech started
+                        is_recording = True
+                        speech_start = time.time()
+                        audio_buffer = []
+                        self.vad_status.emit("recording")
+                        self.status_update.emit("🔴 Recording...")
+                    
+                    # Reset silence counter
+                    silence_start = None
+                    
+                    # Add to buffer
+                    audio_buffer.append(indata.copy())
+                    
+                    # Check max duration
+                    if time.time() - speech_start > max_speech_duration:
+                        # Force stop recording
+                        is_recording = False
+                        self._process_audio(audio_buffer, sample_rate)
+                        audio_buffer = []
+                        self.vad_status.emit("waiting")
+                        
+                else:  # Silence
+                    if is_recording:
+                        # Still recording, add silence chunk
+                        audio_buffer.append(indata.copy())
+                        
+                        # Start or continue silence timer
+                        if silence_start is None:
+                            silence_start = time.time()
+                        
+                        # Check if silence duration exceeded
+                        if time.time() - silence_start > silence_duration:
+                            # Speech ended - process if long enough
+                            speech_duration = time.time() - speech_start
+                            is_recording = False
+                            
+                            if speech_duration >= min_speech_duration:
+                                self._process_audio(audio_buffer, sample_rate)
+                            else:
+                                self.status_update.emit("🎤 (too short, ignored)")
+                            
+                            audio_buffer = []
+                            silence_start = None
+                            self.vad_status.emit("waiting")
+                            self.status_update.emit("🎤 Listening...")
+            
+            # Start audio stream
+            with sd.InputStream(
+                samplerate=sample_rate,
+                channels=1,
+                dtype='float32',
+                blocksize=chunk_samples,
+                callback=audio_callback
+            ):
+                while self._running:
+                    time.sleep(0.1)
+                    
+        except Exception as e:
+            import traceback
+            self.error_occurred.emit(f"Listener error: {e}\n{traceback.format_exc()}")
+        finally:
+            self.vad_status.emit("stopped")
+            self.status_update.emit("🔇 Stopped listening")
+    
+    def _process_audio(self, audio_buffer: list, sample_rate: int):
+        """Process recorded audio - save to file and transcribe"""
+        try:
+            import numpy as np
+            import tempfile
+            import wave
+            import os
+            
+            self.vad_status.emit("processing")
+            self.status_update.emit("⏳ Transcribing...")
+            
+            # Concatenate audio chunks
+            if not audio_buffer:
+                return
+            
+            audio_data = np.concatenate(audio_buffer, axis=0)
+            
+            # Convert to int16
+            audio_int16 = np.int16(audio_data * 32767)
+            
+            # Save to temp file
+            temp_dir = tempfile.gettempdir()
+            temp_path = os.path.join(temp_dir, f"sv_vad_{os.getpid()}.wav")
+            
+            with wave.open(temp_path, 'wb') as wf:
+                wf.setnchannels(1)
+                wf.setsampwidth(2)
+                wf.setframerate(sample_rate)
+                wf.writeframes(audio_int16.tobytes())
+            
+            # Transcribe using API or local model
+            if self.listener.use_api and self.listener.api_key:
+                text = self._transcribe_with_api(temp_path)
+            else:
+                text = self._transcribe_with_local(temp_path)
+            
+            # Clean up
+            try:
+                os.unlink(temp_path)
+            except:
+                pass
+            
+            # Emit result
+            if text:
+                self.transcription_ready.emit(text)
+                
+        except Exception as e:
+            import traceback
+            self.error_occurred.emit(f"Processing error: {e}\n{traceback.format_exc()}")
+
+    def _transcribe_with_api(self, audio_path: str) -> str:
+        """Transcribe using OpenAI Whisper API - much more accurate"""
+        try:
+            from openai import OpenAI
+            
+            client = OpenAI(api_key=self.listener.api_key)
+            
+            with open(audio_path, "rb") as audio_file:
+                # Use whisper-1 model (OpenAI's hosted Whisper)
+                kwargs = {"model": "whisper-1", "file": audio_file}
+                
+                # Add language hint if specified
+                if self.listener.language:
+                    kwargs["language"] = self.listener.language
+                
+                response = client.audio.transcriptions.create(**kwargs)
+            
+            return response.text.strip()
+            
+        except Exception as e:
+            self.error_occurred.emit(f"OpenAI API error: {e}")
+            return ""
+
+    def _transcribe_with_local(self, audio_path: str) -> str:
+        """Transcribe using local Whisper model"""
+        try:
+            if self.listener.language:
+                result = self._model.transcribe(audio_path, language=self.listener.language)
+            else:
+                result = self._model.transcribe(audio_path)
+            
+            return result["text"].strip()
+            
+        except Exception as e:
+            self.error_occurred.emit(f"Local transcription error: {e}")
+            return ""
+
+
+# Legacy class for backwards compatibility
+class _ListenerThread(_VADListenerThread):
+    """Legacy alias for _VADListenerThread"""
+    pass
+