I would appreciate any ideas about improving the program, especially Speaker-function (Automixer).
A +1 for the "Dugan" algorithm. This is public domain so free to use.
Simply put, to perform a Dugan style auto-mix you capture an average level for each channel in the auto-mix, and a average level for the entire mix sum of all (with no gain adjustment). Then compare the individual channel levels to the sum of them all. If a channel is lower level than the sum of all, reduce it by that difference. So if it is .7x the level of the sum apply a gain of .7x, etc. It's that easy,,, This approach delivers a nominal apparent NOM (number of open mics) of 1x.