refactor: audio wake word detector and device capture components

config.yaml

@@ -40,34 +40,34 @@ asr:
   transformers_engine: "huggingface" # "huggingface" or "onnxruntime"
   device: "cpu"  # Pi5 : CPU (or "hailo")
   compute_type: "int8"
-  download_root: null  # Let Python code build the path dynamically
+  download_root: null # Let Python code build the path dynamically
   skip_native_teardown: false  # Avoid native teardown segfaults on some ARM/PortAudio stacks
   store_audio: true
   store_audio_path: ".tmp/asr.wav"
 
 
 wake:
-  wake_word: "hey_jarvis"
-  model_name: "embedding_model"
+  wake_word: "hello_jarvis"
+  backend: "wakeword"
+  model_name: "hey_jarvis_v0.1"
   model_path: null
   inference_framework: "onnx"
   threshold: 0.4
   cooldown_seconds: 2.0
   noise_suppression: true
-  download_root: null
-  vad_threshold: 0.6
+  download_root: ".cache/audio/models/wakeword"
 
 audio:
   input_sample_rate: 48000
-  input_chunk_ms: 30  # taille des chunks audio en ms
-  input_chunk_size: 500
+  input_chunk_ms: 300  # taille des chunks audio en ms
+  input_chunk_size: 1024
   input_device_index: 0 # use system default when a mic is attached
   input_device_name: 'USB ENC Audio Device'  # Optional name of input device to select (overrides index if found)
   volume: 0.5  # half as loud
   output_device_index: 1 # default to USB PnP Audio Device: Audio (hw:3,0)
   output_sample_rate: 44100
-  output_chunk_ms: 30  # taille des chunks audio en ms
-  output_chunk_size: 500
+  output_chunk_ms: 300  # taille des chunks audio en ms
+  output_chunk_size: 1024
   output_device_name: 'USB PnP Audio Device'  # Optional name of output device to select (overrides index if found)
 
 # Platform-specific configuration

examples/VAD/voice_agent_offline.py

@@ -4,7 +4,7 @@
 Minimal voice agent demonstrating ASR, TTS, and wake-word detection.
 
 Features:
-  - Wake word detection (openWakeWord)
+  - Wake word detection (Direct ONNX Runtime, no openwakeword library package)
   - Speech recognition (Whisper or Faster-Whisper)
   - Text-to-speech synthesis (Piper)
   - Simple intent-based response system
@@ -14,6 +14,7 @@
   python examples/simple_voice_agent.py
 """
 
+import datetime
 import logging
 import signal
 import sys
@@ -112,8 +113,15 @@ def _on_wake_word_detected(self) -> None:
         """Callback when wake word is detected."""
         logger.info("🟣 Wake word detected!")
         self.wake_word_active = True
+
+        # Stop listening during TTS playback to avoid hearing ourselves
+        with self._listener_lock:
+            if self._wake_active:
+                self.wake_detector.stop()
+                self._wake_active = False
+
+        self.tts.speak("Yes? How can I help you?", blocking=True)
         self._ensure_asr_mode()
-        self.tts.speak("Yes? How can I help you?")
 
     def _on_transcript_received(self, transcript: str) -> None:
         """Callback when speech is transcribed."""
@@ -131,11 +139,25 @@ def _on_transcript_received(self, transcript: str) -> None:
         # Process the command
         response = self._generate_response(transcript)
         logger.info("🤖 Response: '%s'", response)
-        self.tts.speak(response)
 
-        self.wake_word_active = False
-        logger.info("🟢 Returning to wake word mode")
-        self._ensure_wake_mode()
+        # Stop listening during TTS playback to avoid hearing ourselves
+        with self._listener_lock:
+            if self._asr_active:
+                self.asr.stop()
+                self._asr_active = False
+
+        self.tts.speak(response, blocking=True)
+
+        # Check if we should continue listening in ASR mode or return to wake word mode
+        should_continue = not any(exit_word in transcript.lower() for exit_word in ["stop", "exit", "quit"])
+
+        if should_continue:
+            logger.info("🟢 Continuing conversation, staying in ASR mode")
+            self._ensure_asr_mode()
+        else:
+            self.wake_word_active = False
+            logger.info("🟢 Returning to wake word mode")
+            self._ensure_wake_mode()
 
     def _ensure_wake_mode(self) -> None:
         """Run wake-word listening without a parallel ASR capture stream."""
@@ -184,13 +206,15 @@ def _generate_response(self, user_input: str) -> str:
 
         # Simple keyword matching
         responses = {
-            "hello": "Hello! What can I do for you?",
-            "hi": "Hi there!",
-            "time": "I don't have real-time capabilities right now.",
-            "weather": "I can't check the weather, but I hope it's nice outside!",
-            "help": "I'm a voice agent. Try saying hello or ask me a question.",
-            "thanks": "You're welcome!",
-            "thank you": "Happy to help!",
+            "hello": f"Hello! I'm {self.config.wake.wake_word}, your AI assistant. How can I help you?",
+            "hi": f"Hi there! What can I do for you?",
+            'time': 'The current time is ' + datetime.datetime.now().strftime("%I:%M %p"),  # Get the current time and format it
+            'date': 'Today, the date is: ' + datetime.datetime.now().strftime("%d %B %Y"),  # Get the current date and format it
+            "lights": f"I would control your lights if I had smart home integration.",
+            "music": f"I would play music if I had access to your media system.",
+            "stop": f"Goodbye! Returning to wake word detection.",
+            "bye_bye": f"See you later! Going back to sleep mode.",
+            "help": f"I can respond to simple commands like hello, hi, time, date, lights, music, stop and bye-bye."
         }
 
         # Match keywords

log.json

@@ -17,7 +17,7 @@
     "handlers": {
         "stdout": {
             "class": "logging.StreamHandler",
-            "level": "INFO",
+            "level": "WARNING",
             "filters": ["max_warning"],
             "formatter": "simple",
             "stream": "ext://sys.stdout"

pyproject.toml

@@ -54,8 +54,8 @@ dependencies = [
     "loguru>=0.7.0",
     "numba>=0.57.0",
     "numpy>=1.24.0",
+    "onnxruntime>=1.18.0",
     "openai-whisper==20250625",
-    "onnxruntime>=1.16.0",
     "openwakeword>=0.1.0",
     "piper-tts>=1.4.1",
     "psutil>=7.2.2",
@@ -128,7 +128,7 @@ test = [
 ]
 
 [tool.ruff]
-target-version = "py310"
+target-version = "py311"
 line-length = 120
 output-format = "full"
 src = ["src", "tests", "examples"]

src/audio/asr.py

@@ -6,7 +6,7 @@
 Automatic Speech Recognition engine combining STT + VAD.
 
 Architecture:
-  1. Microphone -> capture thread (PyAudio or sounddevice chunks)
+  1. Microphone -> capture thread sounddevice chunks
   2. Silero VAD -> detect speech vs silence
   3. Accumulate speech chunks
   4. On silence detected -> STT (Whisper/Faster-Whisper) -> transcript
@@ -285,6 +285,13 @@ def start(self, callback: Callable[[str], None]) -> None:
         self._transcript_callback = callback
         self._running = True
 
+        # Clear the queue of any leftover items (including None sentinel)
+        while not self._audio_queue.empty():
+            try:
+                _ = self._audio_queue.get_nowait()
+            except queue.Empty:
+                break
+
         if not self._open_input_stream():
             self._running = False
             return

src/audio/audio_utils.py

@@ -83,14 +83,11 @@ class OpenedInputStream:
 
 def get_audio_backend() -> str:
     try:
-        import sounddevice as sd
-        return "sounddevice"
-    except ImportError:
-        raise ImportError(
-            "The 'sounddevice' library is required but not installed.\n \
-            Install it via: pip install sounddevice"
-        )
-
+        import sounddevice
+        _ = sounddevice
+    except ImportError as e:
+        raise ImportError("The 'sounddevice' library is required") from e
+    return "sounddevice"
 
 def is_backend_available(backend: str) -> bool:
     """Check if a specific audio backend is available.