diff --git a/linux_voice_assistant/__main__.py b/linux_voice_assistant/__main__.py
index 2eed265..764e2be 100644
--- a/linux_voice_assistant/__main__.py
+++ b/linux_voice_assistant/__main__.py
@@ -134,7 +134,7 @@ async def main() -> None:
     finally:
         # --- 10. Cleanup ---
         _LOGGER.debug("Shutting down...")
-        state.mic_muted = True # Signal audio thread to stop
+        state.shutdown = True # Signal audio thread to stop
         audio_thread.join()
         
         if hasattr(state, "mqtt_controller") and state.mqtt_controller:
@@ -404,69 +404,95 @@ def _process_audio(state: ServerState, mic, block_size: int):
     last_active: Optional[float] = None
     
     try:
-        _LOGGER.debug("Opening audio input device: %s", mic.name)
-        with mic.recorder(samplerate=16000, channels=1, blocksize=block_size) as mic_in:
-            while True:
-                if state.mic_muted:
-                    break
-                
-                audio_chunk_array = mic_in.record(block_size).reshape(-1)
-                audio_chunk = (
-                    (np.clip(audio_chunk_array, -1.0, 1.0) * 32767.0)
-                    .astype("<i2")
-                    .tobytes()
-                )
+        while True:
+            # --- THIS IS THE FIX ---
+            # Wait while muted. This loop will spin until unmuted.
+            if state.mic_muted:
+                _LOGGER.debug("Audio thread muted.")
+                while state.mic_muted:
+                    if state.shutdown:
+                        _LOGGER.debug("Shutdown signal received, stopping audio thread.")
+                        return # Exit thread
+                    time.sleep(0.1)
 
-                if state.satellite is None:
-                    time.sleep(0.01)
-                    continue
+                # --- Just unmuted ---
+                _LOGGER.debug("Audio thread unmuted, clearing buffers.")
+                micro_features.reset()
+                if oww_features is not None:
+                    oww_features.reset()
+                last_active = time.monotonic() # Start refractory period
                 
-                # --- THIS IS THE FIX ---
-                if (not wake_words) or (state.wake_words_changed):
-                # --- END FIX ---
-                    state.wake_words_changed = False
-                    wake_words = [
-                        ww for ww in state.wake_words.values() 
-                        if ww.id in state.active_wake_words
-                    ]
-                    has_oww = any(isinstance(ww, OpenWakeWord) for ww in wake_words)
+                # Re-open microphone to flush OS buffer
+                _LOGGER.debug("Re-opening microphone to flush OS buffers...")
+                with mic.recorder(samplerate=16000, channels=1, blocksize=block_size) as mic_in:
+                    mic_in.flush() # Discard any stale data
+                    _LOGGER.debug("Microphone flushed.")
+                    # Fall through to the main recording loop
+            # --- END FIX ---
+            
+            _LOGGER.debug("Opening audio input device: %s", mic.name)
+            with mic.recorder(samplerate=16000, channels=1, blocksize=block_size) as mic_in:
+                # --- This is now the main recording loop ---
+                while not state.mic_muted: 
+                    if state.shutdown:
+                        _LOGGER.debug("Shutdown signal received, stopping audio thread.")
+                        return # Exit thread
                     
-                    if has_oww and (oww_features is None):
-                        _LOGGER.debug("Initializing OpenWakeWord features...")
-                        oww_features = OpenWakeWordFeatures.from_builtin()
-                
-                try:
-                    micro_inputs.clear(); micro_inputs.extend(micro_features.process_streaming(audio_chunk))
+                    audio_chunk_array = mic_in.record(block_size).reshape(-1)
+                    audio_chunk = (
+                        (np.clip(audio_chunk_array, -1.0, 1.0) * 32767.0)
+                        .astype("<i2")
+                        .tobytes()
+                    )
+
+                    if state.satellite is None:
+                        time.sleep(0.01)
+                        continue
                     
-                    if has_oww:
-                        assert oww_features is not None
-                        oww_inputs.clear(); oww_inputs.extend(oww_features.process_streaming(audio_chunk))
-                    
-                    for wake_word in wake_words:
-                        activated = False
-                        if isinstance(wake_word, MicroWakeWord):
-                            if any(wake_word.process_streaming(mi) for mi in micro_inputs): activated = True
-                        elif isinstance(wake_word, OpenWakeWord):
-                            if any(p > 0.5 for oi in oww_inputs for p in wake_word.process_streaming(oi)): activated = True
+                    if (not wake_words) or (state.wake_words_changed):
+                        state.wake_words_changed = False
+                        wake_words = [
+                            ww for ww in state.wake_words.values() 
+                            if ww.id in state.active_wake_words
+                        ]
+                        has_oww = any(isinstance(ww, OpenWakeWord) for ww in wake_words)
                         
-                        if activated:
-                            now = time.monotonic()
-                            if (last_active is None) or ((now - last_active) > state.refractory_seconds):
-                                state.loop.call_soon_threadsafe(state.satellite.wakeup, wake_word)
-                                last_active = now
+                        if has_oww and (oww_features is None):
+                            _LOGGER.debug("Initializing OpenWakeWord features...")
+                            oww_features = OpenWakeWordFeatures.from_builtin()
                     
-                    stopped = False
-                    for micro_input in micro_inputs:
-                        if state.stop_word.process_streaming(micro_input):
-                            stopped = True
-
-                    if stopped and (state.stop_word.id in state.active_wake_words):
-                        state.loop.call_soon_threadsafe(state.satellite.stop)
+                    try:
+                        micro_inputs.clear(); micro_inputs.extend(micro_features.process_streaming(audio_chunk))
                         
-                    state.loop.call_soon_threadsafe(state.satellite.handle_audio, audio_chunk)
+                        if has_oww:
+                            assert oww_features is not None
+                            oww_inputs.clear(); oww_inputs.extend(oww_features.process_streaming(audio_chunk))
+                        
+                        for wake_word in wake_words:
+                            activated = False
+                            if isinstance(wake_word, MicroWakeWord):
+                                if any(wake_word.process_streaming(mi) for mi in micro_inputs): activated = True
+                            elif isinstance(wake_word, OpenWakeWord):
+                                if any(p > 0.5 for oi in oww_inputs for p in wake_word.process_streaming(oi)): activated = True
+                            
+                            if activated:
+                                now = time.monotonic()
+                                if (last_active is None) or ((now - last_active) > state.refractory_seconds):
+                                    state.loop.call_soon_threadsafe(state.satellite.wakeup, wake_word)
+                                    last_active = now
+                        
+                        stopped = False
+                        for micro_input in micro_inputs:
+                            if state.stop_word.process_streaming(micro_input):
+                                stopped = True
 
-                except Exception: 
-                    _LOGGER.exception("Unexpected error handling audio")
+                        if stopped and (state.stop_word.id in state.active_wake_words):
+                            state.loop.call_soon_threadsafe(state.satellite.stop)
+                            
+                        state.loop.call_soon_threadsafe(state.satellite.handle_audio, audio_chunk)
+
+                    except Exception: 
+                        _LOGGER.exception("Unexpected error handling audio")
     
     except Exception as e:
         _LOGGER.critical("A soundcard error occurred: %s", e)