diff --git a/headmic.py b/headmic.py
index 87e01f6..4ad688e 100644
--- a/headmic.py
+++ b/headmic.py
@@ -234,6 +234,7 @@ def listener_loop():
     global state, dual_stream
 
     logger.info("Initializing Porcupine...")
+    porcupine = None
     try:
         porcupine = pvporcupine.create(
             access_key=PORCUPINE_ACCESS_KEY,
@@ -241,8 +242,7 @@ def listener_loop():
         )
     except Exception as e:
         logger.error(f"Failed to init Porcupine: {e}")
-        state.error = str(e)
-        return
+        logger.warning("Wake word detection disabled — audio loop continues for classification")
 
     vad = webrtcvad.Vad(VAD_AGGRESSIVENESS)
 
@@ -264,9 +264,6 @@ def listener_loop():
 
             state.active_side = side
 
-            # Convert bytes to int16 array for Porcupine
-            pcm = struct.unpack_from("h" * 512, frame_data)
-
             # Feed sound classifier ring buffer
             if sound_ring_buffer is not None:
                 sound_ring_buffer.append(frame_data)
@@ -275,8 +272,11 @@ def listener_loop():
             if enrollment_buffer is not None:
                 enrollment_buffer.append(frame_data)
 
-            # Check for wake word
-            keyword_index = porcupine.process(pcm)
+            # Check for wake word (skip if Porcupine not available)
+            keyword_index = -1
+            if porcupine:
+                pcm = struct.unpack_from("h" * 512, frame_data)
+                keyword_index = porcupine.process(pcm)
 
             if keyword_index >= 0 and not is_recording:
                 logger.info("🦊 Wake word detected: 'Hey Vivi'! (from %s ear)", side)
@@ -345,7 +345,8 @@ def listener_loop():
         logger.error(f"Listener error: {e}")
         state.error = str(e)
     finally:
-        porcupine.delete()
+        if porcupine:
+            porcupine.delete()
         state.listening = False
         leds_off()
         logger.info("Listener stopped")