Google recently launched an updated version of the Voice Access. The service allows users to leverage voice commands for control of Android devices. It is based on a machine learning model to automatically identify icons on the screen depending on the UI screenshots. This allows it to analyze if the elements like icons and images have accessibility labels which are provided to accessibility services of Android.
These labels allow the services to refer to one on-screen element at a time and inform the end-users and when they have finished a loop of the UI. Voice Access aims to address some of the elements that lack these labels.