hey, with the new TurnDetection model for Multilingual, when using RealtimeModel from OpenAI, do you still suggest using their VAD or switch to MultiLingualModel? (particularly we relay on semantic vad from openai)

Double checking the external messaging, I don't see any specific recommendations, so take this with a pinch of salt... There's positives and negatives, you should see better accuracy and more consistent performance, especially with non-English speech, and handling structured data (email addresses or phone numbers), but the downside is it would require you to specify a separate STT, which adds cost. I would definitely recommend at least trialling it.