Voice preset is sent as voice_preset. TODO: if backend ignores it, keep current default behavior until API wiring is completed.