fix(voice): decouple streaming asr from media loop

services/ai_voice_runtime_service/app.py

@@ -123,6 +123,10 @@ def _turn_max_ms() -> int:
     return max(_int_env("AI_VOICE_TURN_MAX_MS", 10000), _vad_frame_ms())
 
 
+def _vad_rms_threshold() -> int:
+    return max(_int_env("AI_VOICE_VAD_RMS_THRESHOLD", 250), 1)
+
+
 def _media_idle_timeout_seconds() -> float:
     return max(_float_env("AI_VOICE_MEDIA_IDLE_TIMEOUT_SECONDS", 15.0), 5.0)
 
@@ -1227,6 +1231,7 @@ _MEDIA_RUNTIME = AudioSocketMediaRuntime(
     min_speech_ms=_vad_min_speech_ms(),
     trailing_silence_ms=_vad_trailing_silence_ms(),
     max_turn_ms=_turn_max_ms(),
+    vad_rms_threshold=_vad_rms_threshold(),
     asr_provider=_ASR_PROVIDER,
     streaming_asr_provider=_STREAMING_ASR_PROVIDER,
     tts_provider=_TTS_PROVIDER,