No need for complication.

Your keyboard scan code can be used as is.

But you need to add a second step - postprocessing of the keyboard data based on timing.

When button '2' is pressed, you may display an 'A'.

If no other button is pressed for > x ms, then the 'A' is a keeper.

If a different button is pressed, then the 'A' is a keeper, and you also emit the first possible character for the new button (or accept in case you have any 'enter' or 'ring' or similar button).

If button '2' is returned once more with less than x ms, then you rotate through the possible characters for the '2' button (A, B, C, 2, Å, Ä, ...).
This means that if you have a display where 'A' is visible, you must step back the cursor and replace the 'A' with a 'B'. And next time with a 'C'.

You obviously do need a display, since it is much too complicated for a user to keep track of their input without proper feedback.

In short - the keyboard scan code can be unchanged. It's enough that it returns which of the 12 keys that is pressed, after having properly debounced the keyboard. It's the layer above (probably a function called from the main loop) that gets a bit more complicated, but not by much.