Twilio Changelog | Nov. 13, 2026
General Availability: Smart Turn Detection for Conversation Relay
Eliminate unnatural mid-sentence AI interruptions with the General Availability of Smart Turn Detection for Conversation Relay. Our new in-house, semantics-aware model evaluates speech meaning in real time to accurately detect when a caller has actually finished speaking, replacing rigid silence timers for a much more natural conversational experience.
What’s New:
Semantics-Aware End-of-Turn: The system now distinguishes between a caller who is finished speaking and a caller who is simply pausing mid-sentence to think or use filler words, eliminating unnatural AI interruptions and "choppy" dialogue.
Universal STT Support: Unlike previous models that locked you into a specific provider, this in-house engine runs natively in Twilio’s media stack (TME) and works across all supported STT providers (excluding Deepgram Flux, which bypasses Twilio's engine to use its own built-in turn detection).
New TwiML Controls: Easily toggle the feature on or off at the call level by adding the turnDetection="true" parameter to your <ConversationRelay> TwiML element.
Enriched WebSocket Events: Developers can add includeInPrompt="turnState" to their TwiML to receive explicit turn metadata (StartOfTurn, EagerEndOfTurn, EndOfTurn) directly in their WebSocket payloads for granular application-level orchestration.
Get Started: Turn Detection is now Generally Available. Check out the docs to learn how to leverage intelligent, semantics-aware turn taking and deliver more natural, interruption-free voice AI experiences today.