videosphere
    Preparing search index...
    • Maps an AssemblyAI Turn frame to a soft-split kind for caption emission.

      With format_turns=true, AssemblyAI sends an unformatted end_of_turn then a formatted one — treating both as provider finals would duplicate captions. Soft-split still runs on the unformatted turn as a partial so long speech becomes sentence-sized finals before the formatted flush.

      Parameters

      • msg: { end_of_turn?: boolean; turn_is_formatted?: boolean }

        Parsed Turn fields.

      Returns "partial" | "final"

      final when the turn is complete (formatted if formatting is on).