The voice input interaction has its own controls. A canonical record button, a full talk-mode wrapper, and a live waveform that fills its container, built on the same shared primitives as the rest of the UI, so the affordance reads the same in every screen you open it.