TranscriptTurn

What was said, as text.

Who spoke this turn — the Resia agent (assistant), the person it called (user), the system, or a non-speech event such as a keypad tone.

Deprecated: use start_secs to place the turn in the audio and started_at for the wall-clock instant. Unix epoch seconds at which the pipeline recorded the turn, which is not the moment speech started.

How spoken_text was determined: exact-pts (an exact word-timed capture), estimate-16cps or sentence-sync (estimates), or none.

Why an assistant turn was cut short — the caller talked over it (barge-in), or the call ended.

When the turn ends, in seconds, on the same scale and from the same zero as start_secs.

True when an assistant turn was cut off before it finished speaking.

Who was on the line when this turn was spoken, as participant names. The agent is implicit and the speaker is always a member. Absent means every participant declared on the call was on the line; an empty list means everyone was on hold and the agent was momentarily alone. A person heard a turn exactly when this field is absent or contains their name.

Which human spoke this turn, as a participant name from the Call Agent's roster — the primary party uses the agent's primary_participant name. Present on every human turn of a multi-party call; absent on a plain two-party call.

For an interrupted assistant turn, how many seconds of it were actually spoken before the cut.

For an interrupted assistant turn, the portion actually spoken before the cut, when known.

When the turn starts, in seconds from the first sample of the call's recording. The count starts at the first inbound audio frame when the call has no recording. Null on multi-party calls and on turns whose position could not be measured.

When the turn started, as an instant. Derived from the recording's start plus start_secs; approximate.

The pipeline's identifier for this turn; empty when it was not set.

Per-word timings for this turn, when the transcriber captured them; empty otherwise.