Enhance DuplexPipeline and AssistantsPage for improved interruption handling. Introduce _OPENER_PRE_ROLL_MS constant for a head start on mic capture, and adjust interruption sensitivity settings from 500ms to 180ms across relevant components to optimize responsiveness during assistant interactions.

engine/core/duplex_pipeline.py

@@ -73,6 +73,7 @@ class DuplexPipeline:
     _ASR_DELTA_THROTTLE_MS = 500
     _LLM_DELTA_THROTTLE_MS = 80
     _ASR_CAPTURE_MAX_MS = 15000
+    _OPENER_PRE_ROLL_MS = 180
     _DEFAULT_TOOL_SCHEMAS: Dict[str, Dict[str, Any]] = {
         "current_time": {
             "name": "current_time",
@@ -781,6 +782,9 @@ class DuplexPipeline:
         )
         await self.conversation.add_assistant_turn(greeting_to_speak)
 
+        # Give client mic capture a short head start so opener can be interrupted immediately.
+        await asyncio.sleep(self._OPENER_PRE_ROLL_MS / 1000.0)
+
         used_preloaded_audio = await self._play_preloaded_opener_audio()
         if self._tts_output_enabled() and not used_preloaded_audio:
             # Keep opener text ahead of opener voice start.