Hey all! Any opinions on what VAD (voice activity detection) parameters work best for smooth conversation flow (no awkward interruptions) in a phone call setting? with the nova3-multi + gpt4.1-mini + cartesia-sonic2 combo

Set waitSeconds to 0.6–1.2s, enable smartEndpointing, use onPunctuationSeconds at 0.5s, onNoPunctuationSeconds at 1.5s, and adjust stop-speak VAD with numWords at 2–3, voiceSeconds at 0.3–0.5s, and backoffSeconds at 1–1.5s, while also enabling input reformatting and preprocessing for Sonic2.

Have you had any success with smart endpointing from livekit? any optimal parameter settings for phone calls?

LiveKit smart endpointing works well for phone calls when using the smart mode with tuned parameters like a 0.5s–6s delay range, a 500ms silence threshold, and optional noise cancellation for better speech detection.