How does the algorithm ensure you’re not amplifying the loudest voice?

How does the algorithm ensure you’re not amplifying the loudest voice?