In my case, I had speakers with the cone mounted using a concave rubber ring (most rubber-hinged speakers have a convex ring). I glued the mirror on this rubber ring, so a movement of the cone would move one side of the mirror up or down almost as much as the cone, while the other side would hardly move at all.

            +---- mirror
            v
         -====/
        /  --/ <--- rubber
       /
      / <--- stiff cone
     /
    /
---/

But another way people have done it is like (normally with an almost flat speaker):

   +----------- mechanical arm, that gets tilted by the movement of the center of the cone.
   |   +------- mirror
   v   v
 +----===-* <--- hinge of rubber, cloth or silicon glue or similar
 |      /
 |     /
 |    /
 |   /
-*--/

When drawing a graph, it isn't the absolute errors that you will notice. It is the repeatability. One of the photos in the link showed the repeatability with the lower alternative. My implementation managed a bit better, but it is impossible to say if it was because of luck, the difference in hw design or the difference in how I controlled the deflection.

Anyway, the link I posted contains a quite good photo of the second mounting alternative. I haven't seen anyone else try the alternative I used. Most probably because of the availability of a suitable loudspeaker - it is more common that the rubbber ring is convex. And small and cheap elements have a glued cone. It was also a bit of a problem fixing the mirror - rubber and fat plastics are always a bit of a problem.