fix: increase STT silence_duration_ms to prevent choppy transcription

Default silence_duration_ms=350 is too aggressive for Chinese speech,
causing sentences to be fragmented into 1-3 character chunks. Increase
to 800ms and raise VAD threshold to 0.6 so the STT waits longer before
finalizing a turn, producing complete sentences for LLM processing.

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

packages/services/voice-agent/src/agent.py

@@ -203,6 +203,14 @@ async def entrypoint(ctx: JobContext) -> None:
             language=settings.whisper_language,
             client=_oai_client,
             use_realtime=True,
+            # Increase silence_duration_ms so Chinese speech isn't chopped
+            # into tiny fragments (default 350ms is too aggressive).
+            turn_detection={
+                "type": "server_vad",
+                "threshold": 0.6,
+                "prefix_padding_ms": 600,
+                "silence_duration_ms": 800,
+            },
         )
     else:
         stt = LocalWhisperSTT(