voice-mode 2.28.2__tar.gz → 2.29.0__tar.gz

{voice_mode-2.28.2 → voice_mode-2.29.0}/CHANGELOG.md

@@ -7,14 +7,60 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.29.0] - 2025-08-25
+
+### Added
+- **CoreML acceleration support for Whisper on Apple Silicon**
+  - Added optional dependency group 'coreml' with PyTorch and CoreMLTools
+  - Enhanced whisper_model_install tool with install_torch and auto_confirm parameters
+  - Automatic detection of Apple Silicon Macs with CoreML acceleration offer
+  - User-friendly confirmation prompts for large (~2.5GB) PyTorch download
+  - Graceful fallback to Metal acceleration if CoreML requirements not met
+  - Clear instructions for enabling CoreML later if initially skipped
+
+- **Beautiful installer experience**
+  - Added Voice Mode ASCII art in Claude Code orange color
+  - Enhanced preamble with clear value proposition and privacy messaging
+  - Early system detection with special recognition for Apple Silicon
+  - Professional presentation with centered text and visual hierarchy
+
+### Fixed
+- **Improved converse tool documentation**
+  - Simplified listen_duration parameter documentation
+  - Removed confusing duration recommendations that led to unnecessary overrides
+  - Clarified that silence detection handles timing well with sensible defaults
+  - Reduces cognitive load and prevents token waste from explicit duration settings
+
+## [2.28.3] - 2025-08-24
+
+### Fixed
+- **Parameter type handling for MCP tools**
+  - Fixed vad_aggressiveness parameter to accept string values from LLMs
+  - Fixed port parameters in kokoro_install and livekit_install
+  - Fixed lines parameter in service management tool
+  - All numeric parameters now properly convert strings to integers
+  - Addresses systemic issue where Claude Code MCP client passes strings
+
+- **Installer script uvx command corrections**
+  - Fixed MCP configuration to use correct command `uvx voice-mode` (without --refresh)
+  - Installer now always refreshes to latest version at start
+  - Removed unnecessary --refresh flags from runtime commands
+  - Updated user-facing command examples to show correct usage
+
 ## [2.28.2] - 2025-08-24
 
+### Added
+- **Configurable audio feedback pip delays**
+  - Added VOICEMODE_PIP_LEADING_SILENCE and VOICEMODE_PIP_TRAILING_SILENCE environment variables
+  - Allows customization of silence before and after audio feedback chimes
+  - Configurable via converse tool parameters pip_leading_silence and pip_trailing_silence
+  - Helps prevent audio cutoff on Bluetooth devices and other audio systems with delay
+
 ### Fixed
-- **Improved noise filtering in VAD (Voice Activity Detection)**
-  - Fixed issue where nose blowing and similar non-speech sounds were detected as speech
-  - Adjusted VAD aggressiveness handling to better filter continuous non-speech audio
-  - Prevents false positives from breathing sounds, sniffles, and ambient noise
-  - Improves overall speech detection accuracy
+- **Audio feedback for Bluetooth devices**
+  - Added silence buffer before chimes to prevent Bluetooth audio cutoff
+  - Improved compatibility with devices that have audio activation delay
+  - Better audio feedback experience across different output devices
 
 ## [2.28.1] - 2025-08-24
 

{voice_mode-2.28.2 → voice_mode-2.29.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: voice-mode
-Version: 2.28.2
+Version: 2.29.0
 Summary: VoiceMode - Voice interaction capabilities for AI assistants (formerly voice-mcp)
 Project-URL: Homepage, https://github.com/mbailey/voicemode
 Project-URL: Repository, https://github.com/mbailey/voicemode
@@ -39,6 +39,11 @@ Requires-Dist: simpleaudio
 Requires-Dist: sounddevice
 Requires-Dist: uv>=0.4.0
 Requires-Dist: webrtcvad>=2.0.10
+Provides-Extra: coreml
+Requires-Dist: ane-transformers; extra == 'coreml'
+Requires-Dist: coremltools>=7.0; extra == 'coreml'
+Requires-Dist: torch>=2.0.0; extra == 'coreml'
+Requires-Dist: transformers; extra == 'coreml'
 Provides-Extra: dev
 Requires-Dist: build>=1.0.0; extra == 'dev'
 Requires-Dist: pytest-asyncio>=0.21.0; extra == 'dev'

{voice_mode-2.28.2 → voice_mode-2.29.0}/pyproject.toml

@@ -50,6 +50,12 @@ dependencies = [
 ]
 
 [project.optional-dependencies]
+coreml = [
+    "torch>=2.0.0",
+    "coremltools>=7.0",
+    "transformers",
+    "ane-transformers",
+]
 dev = [
     "build>=1.0.0",
     "twine>=4.0.0",

{voice_mode-2.28.2 → voice_mode-2.29.0}/voice_mode/__version__.py

@@ -1,3 +1,3 @@
 # This file is automatically updated by 'make release'
 # Do not edit manually
-__version__ = "2.28.2"
+__version__ = "2.29.0"

{voice_mode-2.28.2 → voice_mode-2.29.0}/voice_mode/tools/converse.py

@@ -1323,7 +1323,7 @@ async def converse(
     audio_format: Optional[str] = None,
     disable_silence_detection: Union[bool, str] = False,
     speed: Optional[float] = None,
-    vad_aggressiveness: Optional[int] = None,
+    vad_aggressiveness: Optional[Union[int, str]] = None,
     skip_tts: Optional[Union[bool, str]] = None,
     pip_leading_silence: Optional[float] = None,
     pip_trailing_silence: Optional[float] = None
@@ -1351,14 +1351,13 @@ async def converse(
         message: The message to speak
         wait_for_response: Whether to listen for a response after speaking (default: True)
         listen_duration: How long to listen for response in seconds (default: 120.0)
-                         Recommended durations based on expected response:
-                         - Simple yes/no questions: 10 seconds
-                         - Normal conversational responses: 30 seconds  
-                         - Open-ended questions: 60 seconds
-                         - Detailed explanations: 120 seconds (default)
-                         - Stories or long explanations: 300 seconds
-                         Always err on the side of longer duration - it's better to have 
-                         silence at the end than to cut off the user mid-sentence.
+                         The tool handles silence detection well and uses a sensible default.
+                         It's unusual to need to set the duration - only override if you have 
+                         specific requirements such as:
+                         - Silence detection is disabled and you need a specific timeout
+                         - You know the response will be exceptionally long (>120s)
+                         - You're in a special mode that requires different timing
+                         In most cases, just let the default and silence detection handle it.
         min_listen_duration: Minimum time to record before silence detection can stop (default: 2.0)
                              Useful for preventing premature cutoffs when users need thinking time.
                              Examples:
@@ -1495,6 +1494,15 @@ async def converse(
     if skip_tts is not None and isinstance(skip_tts, str):
         skip_tts = skip_tts.lower() in ('true', '1', 'yes', 'on')
     
+    # Convert vad_aggressiveness to integer if provided as string
+    if vad_aggressiveness is not None and isinstance(vad_aggressiveness, str):
+        try:
+            vad_aggressiveness = int(vad_aggressiveness)
+            # Validation will happen later in the function
+        except ValueError:
+            logger.warning(f"Invalid VAD aggressiveness value '{vad_aggressiveness}', using default")
+            vad_aggressiveness = None
+    
     # Determine whether to skip TTS
     if skip_tts is not None:
         # Parameter explicitly set, use it

{voice_mode-2.28.2 → voice_mode-2.29.0}/voice_mode/tools/service.py

@@ -8,7 +8,7 @@ import platform
 import subprocess
 import time
 from pathlib import Path
-from typing import Literal, Optional, Dict, Any
+from typing import Literal, Optional, Dict, Any, Union
 
 import psutil
 
@@ -980,7 +980,7 @@ async def view_logs(service_name: str, lines: Optional[int] = None) -> str:
 async def service(
     service_name: Literal["whisper", "kokoro", "livekit", "frontend"],
     action: Literal["status", "start", "stop", "restart", "enable", "disable", "logs", "update-service-files"] = "status",
-    lines: Optional[int] = None
+    lines: Optional[Union[int, str]] = None
 ) -> str:
     """Unified service management tool for voice mode services.
     
@@ -1007,6 +1007,14 @@ async def service(
         service("kokoro", "start")    # Start Kokoro service
         service("whisper", "logs", 100)  # View last 100 lines of Whisper logs
     """
+    # Convert lines to integer if provided as string
+    if lines is not None and isinstance(lines, str):
+        try:
+            lines = int(lines)
+        except ValueError:
+            logger.warning(f"Invalid lines value '{lines}', using default 50")
+            lines = 50
+    
     # Route to appropriate handler
     if action == "status":
         return await status_service(service_name)

{voice_mode-2.28.2 → voice_mode-2.29.0}/voice_mode/tools/services/kokoro/install.py

@@ -26,7 +26,7 @@ logger = logging.getLogger("voice-mode")
 async def kokoro_install(
     install_dir: Optional[str] = None,
     models_dir: Optional[str] = None,
-    port: int = 8880,
+    port: Union[int, str] = 8880,
     auto_start: Union[bool, str] = True,
     install_models: Union[bool, str] = True,
     force_reinstall: Union[bool, str] = False,
@@ -54,6 +54,14 @@ async def kokoro_install(
         Installation status with service configuration details
     """
     try:
+        # Convert port to integer if provided as string
+        if isinstance(port, str):
+            try:
+                port = int(port)
+            except ValueError:
+                logger.warning(f"Invalid port value '{port}', using default 8880")
+                port = 8880
+        
         # Check for and migrate old installations
         migration_msg = auto_migrate_if_needed("kokoro")
         

{voice_mode-2.28.2 → voice_mode-2.29.0}/voice_mode/tools/services/livekit/install.py

@@ -124,7 +124,7 @@ room:
 @mcp.tool()
 async def livekit_install(
     install_dir: Optional[str] = None,
-    port: int = 7880,
+    port: Union[int, str] = 7880,
     force_reinstall: Union[bool, str] = False,
     auto_enable: Optional[Union[bool, str]] = None,
     version: str = "latest"
@@ -159,6 +159,14 @@ async def livekit_install(
         # Check system
         system = platform.system()
         
+        # Convert port to integer if provided as string
+        if isinstance(port, str):
+            try:
+                port = int(port)
+            except ValueError:
+                logger.warning(f"Invalid port value '{port}', using default 7880")
+                port = 7880
+        
         # Handle string boolean conversions
         if isinstance(force_reinstall, str):
             force_reinstall = force_reinstall.lower() in ("true", "1", "yes", "on")

{voice_mode-2.28.2 → voice_mode-2.29.0}/voice_mode/tools/services/whisper/model_install.py

@@ -1,10 +1,13 @@
 """Download Whisper models with Core ML support."""
 
 import os
+import sys
 import json
 import logging
+import platform
+import subprocess
 from pathlib import Path
-from typing import Union, List
+from typing import Union, List, Dict, Any
 
 from voice_mode.server import mcp
 from voice_mode.config import logger, MODELS_DIR
@@ -17,7 +20,9 @@ logger = logging.getLogger("voice-mode")
 async def whisper_model_install(
     model: Union[str, List[str]] = "large-v2",
     force_download: Union[bool, str] = False,
-    skip_core_ml: Union[bool, str] = False
+    skip_core_ml: Union[bool, str] = False,
+    install_torch: Union[bool, str] = False,
+    auto_confirm: Union[bool, str] = False
 ) -> str:
     """Download Whisper model(s) with optional Core ML conversion.
     
@@ -31,6 +36,8 @@ async def whisper_model_install(
                - "all" to download all available models
         force_download: Re-download even if model exists (default: False)
         skip_core_ml: Skip Core ML conversion on Apple Silicon (default: False)
+        install_torch: Install PyTorch for CoreML (adds ~2.5GB) (default: False)
+        auto_confirm: Skip all confirmation prompts (default: False)
         
     Available models:
         - tiny, tiny.en
@@ -77,6 +84,20 @@ async def whisper_model_install(
                 "error": "Whisper.cpp not installed. Please run whisper_install first."
             }, indent=2)
         
+        # Handle CoreML dependencies if needed
+        coreml_status = await _handle_coreml_dependencies(
+            install_torch=install_torch,
+            auto_confirm=auto_confirm,
+            skip_core_ml=skip_core_ml
+        )
+        
+        if not coreml_status["continue"]:
+            return json.dumps(coreml_status, indent=2)
+        
+        # If CoreML deps were installed, skip_core_ml may have been updated
+        if coreml_status.get("coreml_deps_failed"):
+            skip_core_ml = True
+        
         # Parse model input
         available_models = get_available_models()
         
@@ -200,4 +221,79 @@ async def whisper_model_install(
         return json.dumps({
             "success": False,
             "error": str(e)
-        }, indent=2)
+        }, indent=2)
+
+
+async def _handle_coreml_dependencies(
+    install_torch: bool = False,
+    auto_confirm: bool = False,
+    skip_core_ml: bool = False
+) -> Dict[str, Any]:
+    """Handle CoreML dependency installation for Apple Silicon Macs.
+    
+    Returns:
+        Dict with 'continue' key indicating whether to proceed with model download
+    """
+    # Check if we're on Apple Silicon Mac
+    if platform.system() != "Darwin" or platform.machine() != "arm64":
+        return {"continue": True}
+    
+    # If skipping CoreML, no need to check dependencies
+    if skip_core_ml:
+        return {"continue": True}
+    
+    # Check if torch is already installed
+    try:
+        import torch
+        logger.info("PyTorch already installed for CoreML support")
+        return {"continue": True}
+    except ImportError:
+        pass
+    
+    # Check if user wants to install torch
+    if not install_torch and not auto_confirm:
+        return {
+            "continue": False,
+            "success": False,
+            "requires_confirmation": True,
+            "message": "CoreML requires PyTorch (~2.5GB). Rerun with install_torch=True to confirm.",
+            "recommendation": "Set install_torch=True for CoreML acceleration (2-3x faster)"
+        }
+    
+    # Install CoreML dependencies
+    logger.info("Installing CoreML dependencies...")
+    
+    try:
+        # Detect environment and install appropriately
+        packages = ["torch>=2.0.0", "coremltools>=7.0", "transformers", "ane-transformers"]
+        
+        # Try UV first (most common)
+        if subprocess.run(["which", "uv"], capture_output=True).returncode == 0:
+            cmd = ["uv", "pip", "install"] + packages
+            logger.info("Installing via UV...")
+        else:
+            # Fallback to pip
+            cmd = [sys.executable, "-m", "pip", "install"] + packages
+            logger.info("Installing via pip...")
+        
+        # Run installation
+        result = subprocess.run(cmd, capture_output=True, text=True)
+        
+        if result.returncode == 0:
+            logger.info("CoreML dependencies installed successfully")
+            return {"continue": True, "coreml_deps_installed": True}
+        else:
+            logger.warning(f"Failed to install CoreML dependencies: {result.stderr}")
+            return {
+                "continue": True,
+                "coreml_deps_failed": True,
+                "warning": "CoreML dependencies installation failed. Models will use Metal acceleration."
+            }
+            
+    except Exception as e:
+        logger.warning(f"Error installing CoreML dependencies: {e}")
+        return {
+            "continue": True,
+            "coreml_deps_failed": True,
+            "warning": f"CoreML setup error: {str(e)}. Models will use Metal acceleration."
+        }