Terminology

• A turn is a single spoken segment from the user. In the current API you mark a turn explicitly, sending start_turn before the user speaks and end_turn when they are done. Each turn is processed and transcribed by the speech recognition engine. Turn partials are incremental results emitted during the turn (the turn_partial events), containing the transcript decoded so far.
• A session is the full duration of a WebSocket connection. It begins with start_session and typically contains multiple turns as the user and agent interact back and forth.
• Turn-taking mode refers to how turn boundaries are controlled. The current production API supports manual mode, where the client explicitly controls both the start (via start_turn) and end (via end_turn) of each turn. Automatic turn-taking is on the roadmap.

See the High-Level Flow for how these concepts fit together in a session, or the WebSocket API Reference for the exact messages.