fix(vad): raise activation threshold to reduce noise triggers

activation_threshold 0.5→0.75, min_speech_duration 0.05→0.2s
Prevents ambient noise from triggering STT and producing
'Schlechte Qualität' transcripts.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

voice.py

@@ -59,7 +59,11 @@ _vad = None
 def _get_vad():
     global _vad
     if _vad is None:
-        _vad = silero.VAD.load()
+        _vad = silero.VAD.load(
+            activation_threshold=0.75,
+            min_speech_duration=0.2,
+            min_silence_duration=0.55,
+        )
     return _vad
 
 def _make_lk_identity(user_id, device_id):