output.volume. Set it when you create or update the agent, or inline over the WebSocket via session.update.
Accepts a number from 0 (silent) to 100 (loudest). If omitted, the voice plays at its native level.
Unlike
output.voice and output.format, which are immutable after session.ready, output.volume can be changed mid-session via session.update. The new value applies to subsequent reply.audio chunks.