Hi Bojcha,
The singleband compressor currently does NOT use the RMS volume. It doesn't work that badly, but it must be possible to let it work much better.
I've now (not yet uploaded!) added an RMS calculation step, so for each portion of sound I know the RMS volume. What remains is reducing the volume at spikes, such that I eventually reach the wanted (constant) RMS volume. It's a bit more difficult than I had expected (and hoped), but it should be possible.
My first attempt was to use the peak level and the RMS volume, which worked remarkably well - except that for some sounds (basically female voices / trumpet / strings like sounds) the peak level is very high because the signal is not symmetrical. And then the RMS/peak factor is too low (RMS volume compared to peak level seems to be very low), which in turn leads to a much higher output level

So 95% of the sounds are within a 1 dB range (even if I compare your very badly mixed song with 1980s tracks which have almost no compression, which is pretty good), but some are a few dB louder, which is far too much.
I hope to get something working tomorrow...