voice-mode 2.19.0__tar.gz → 2.20.1__tar.gz

{voice_mode-2.19.0 → voice_mode-2.20.1}/CHANGELOG.md

@@ -7,8 +7,49 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [2.20.1] - 2025-08-11
+
+### Fixed
+- **Speed parameter validation error** - Fixed MCP validation error when passing speed parameter as string
+  - Added type conversion from string to float for speed parameter
+  - Now properly handles speed values passed by MCP clients (e.g., via uvx)
+  - Added comprehensive validation and error messages for invalid speed values
+
+## [2.20.0] - 2025-08-10
+
+### Added
+- **VAD aggressiveness control**
+  - New `vad_aggressiveness` parameter in converse tool for controlling Voice Activity Detection sensitivity (0-3)
+  - 0 = least aggressive filtering (more permissive), 3 = most aggressive (strict)
+  - Allows adapting to different environments: quiet rooms (0-1) vs noisy environments (2-3)
+  - Also configurable via VOICEMODE_VAD_AGGRESSIVENESS environment variable
+
+### Changed
+- **Improved VAD documentation**
+  - Clarified that aggressiveness controls how strictly VAD filters out non-speech
+  - Updated examples to better demonstrate appropriate use cases
+  - Fixed configuration documentation that had backwards descriptions
+
 ## [2.19.0] - 2025-08-10
 
+### Added
+- **MCP prompt command: /release-notes**
+  - New command to display recent changelog entries directly in Claude Code
+  - Shows 5 most recent versions by default (configurable with parameter)
+  - Parses and formats CHANGELOG.md for easy reading
+  - Inspired by Claude Code's own /release-notes feature
+  - Includes comprehensive test coverage
+
+### Fixed
+- Release notes prompt now handles empty string parameters correctly
+- Command works properly with both source and installed packages
+- Changelog is now accessible as an MCP resource when package is installed
+
+### Changed
+- Release notes output format now matches Claude Code's clean, minimal style
+- Removed decorative headers and footers for cleaner terminal output
+- Release notes displayed in chronological order (oldest first)
+
 ## [2.18.0] - 2025-08-10
 
 ### Added

{voice_mode-2.19.0 → voice_mode-2.20.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: voice-mode
-Version: 2.19.0
+Version: 2.20.1
 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

{voice_mode-2.19.0 → voice_mode-2.20.1}/voice_mode/__version__.py

@@ -1,3 +1,3 @@
 # This file is automatically updated by 'make release'
 # Do not edit manually
-__version__ = "2.19.0"
+__version__ = "2.20.1"

{voice_mode-2.19.0 → voice_mode-2.20.1}/voice_mode/tools/converse.py

@@ -871,7 +871,7 @@ def record_audio(duration: float) -> np.ndarray:
             sys.stderr = original_stderr
 
 
-def record_audio_with_silence_detection(max_duration: float, disable_silence_detection: bool = False, min_duration: float = 0.0) -> np.ndarray:
+def record_audio_with_silence_detection(max_duration: float, disable_silence_detection: bool = False, min_duration: float = 0.0, vad_aggressiveness: Optional[int] = None) -> np.ndarray:
     """Record audio from microphone with automatic silence detection.
     
     Uses WebRTC VAD to detect when the user stops speaking and automatically
@@ -881,6 +881,7 @@ def record_audio_with_silence_detection(max_duration: float, disable_silence_det
         max_duration: Maximum recording duration in seconds
         disable_silence_detection: If True, disables silence detection and uses fixed duration recording
         min_duration: Minimum recording duration before silence detection can stop (default: 0.0)
+        vad_aggressiveness: VAD aggressiveness level (0-3). If None, uses VAD_AGGRESSIVENESS from config
         
     Returns:
         Numpy array of recorded audio samples
@@ -902,8 +903,9 @@ def record_audio_with_silence_detection(max_duration: float, disable_silence_det
     logger.info(f"🎤 Recording with silence detection (max {max_duration}s)...")
     
     try:
-        # Initialize VAD
-        vad = webrtcvad.Vad(VAD_AGGRESSIVENESS)
+        # Initialize VAD with provided aggressiveness or default
+        effective_vad_aggressiveness = vad_aggressiveness if vad_aggressiveness is not None else VAD_AGGRESSIVENESS
+        vad = webrtcvad.Vad(effective_vad_aggressiveness)
         
         # Calculate chunk size (must be 10, 20, or 30ms worth of samples)
         chunk_samples = int(SAMPLE_RATE * VAD_CHUNK_DURATION_MS / 1000)
@@ -932,7 +934,7 @@ def record_audio_with_silence_detection(max_duration: float, disable_silence_det
         original_stdout = sys.stdout
         original_stderr = sys.stderr
         
-        logger.debug(f"VAD config - Aggressiveness: {VAD_AGGRESSIVENESS}, "
+        logger.debug(f"VAD config - Aggressiveness: {effective_vad_aggressiveness} (param: {vad_aggressiveness}, default: {VAD_AGGRESSIVENESS}), "
                     f"Silence threshold: {SILENCE_THRESHOLD_MS}ms, "
                     f"Min duration: {MIN_RECORDING_DURATION}s, "
                     f"Initial grace period: {INITIAL_SILENCE_GRACE_PERIOD}s")
@@ -1227,7 +1229,8 @@ async def converse(
     audio_feedback_style: Optional[str] = None,
     audio_format: Optional[str] = None,
     disable_silence_detection: Union[bool, str] = False,
-    speed: Optional[float] = None
+    speed: Optional[float] = None,
+    vad_aggressiveness: Optional[int] = None
 ) -> str:
     """Have a voice conversation - speak a message and optionally listen for response.
     
@@ -1289,6 +1292,16 @@ async def converse(
         speed: Speech rate/speed for TTS playback (default: None uses normal speed)
                Values: 0.25 to 4.0 (0.5 = half speed, 2.0 = double speed)
                Supported by both OpenAI and Kokoro TTS providers.
+        vad_aggressiveness: Voice Activity Detection aggressiveness level (default: None uses VOICEMODE_VAD_AGGRESSIVENESS env var)
+                            Controls how strict the VAD is about filtering out non-speech audio.
+                            Values: 0-3 (integer)
+                            - 0: Least aggressive filtering - includes more audio, may include non-speech
+                            - 1: Slightly stricter filtering
+                            - 2: Balanced filtering (default) - good for most environments
+                            - 3: Most aggressive filtering - strict speech detection, may cut off soft speech
+                            
+                            Use lower values (0-1) in quiet environments to catch all speech
+                            Use higher values (2-3) in noisy environments to reduce false triggers
         If wait_for_response is False: Confirmation that message was spoken
         If wait_for_response is True: The voice response received (or error/timeout message)
     
@@ -1320,6 +1333,15 @@ async def converse(
         - Slower speech: converse("This is slower speech", speed=0.8)
         
         Note: Speed control works with both OpenAI and Kokoro TTS providers
+    
+    VAD Aggressiveness Examples:
+        - Quiet room, capture all speech: converse("Let's have a conversation", vad_aggressiveness=0)
+        - Normal home/office: converse("Tell me about your day")  # Uses default (2)
+        - Noisy cafe/outdoors: converse("Can you hear me?", vad_aggressiveness=3)
+        - Balance for most cases: converse("How are you?", vad_aggressiveness=2)
+        
+        Remember: Lower values (0-1) = more permissive, may detect non-speech as speech
+                 Higher values (2-3) = more strict, may miss soft speech or whispers
     """
     # Convert string booleans to actual booleans
     if isinstance(wait_for_response, str):
@@ -1329,8 +1351,25 @@ async def converse(
     if isinstance(audio_feedback, str):
         audio_feedback = audio_feedback.lower() in ('true', '1', 'yes', 'on')
     
+    # Convert string speed to float
+    if speed is not None and isinstance(speed, str):
+        try:
+            speed = float(speed)
+        except ValueError:
+            return f"❌ Error: speed must be a number (got '{speed}')"
+    
+    # Validate speed parameter range
+    if speed is not None:
+        if not (0.25 <= speed <= 4.0):
+            return f"❌ Error: speed must be between 0.25 and 4.0 (got {speed})"
+    
     logger.info(f"Converse: '{message[:50]}{'...' if len(message) > 50 else ''}' (wait_for_response: {wait_for_response})")
     
+    # Validate vad_aggressiveness parameter
+    if vad_aggressiveness is not None:
+        if not isinstance(vad_aggressiveness, int) or vad_aggressiveness < 0 or vad_aggressiveness > 3:
+            return f"Error: vad_aggressiveness must be an integer between 0 and 3 (got {vad_aggressiveness})"
+    
     # Validate duration parameters
     if wait_for_response:
         if min_listen_duration < 0:
@@ -1604,9 +1643,9 @@ async def converse(
                         event_logger.log_event(event_logger.RECORDING_START)
                     
                     record_start = time.perf_counter()
-                    logger.debug(f"About to call record_audio_with_silence_detection with duration={listen_duration}, disable_silence_detection={disable_silence_detection}, min_duration={min_listen_duration}")
+                    logger.debug(f"About to call record_audio_with_silence_detection with duration={listen_duration}, disable_silence_detection={disable_silence_detection}, min_duration={min_listen_duration}, vad_aggressiveness={vad_aggressiveness}")
                     audio_data = await asyncio.get_event_loop().run_in_executor(
-                        None, record_audio_with_silence_detection, listen_duration, disable_silence_detection, min_listen_duration
+                        None, record_audio_with_silence_detection, listen_duration, disable_silence_detection, min_listen_duration, vad_aggressiveness
                     )
                     timings['record'] = time.perf_counter() - record_start