The DSP does also tricks to make them "louder than possible". That's why it was complicated for the Asahi team : messing with the algo would easily fry the speakers because they're driven by more current than they can sustain, so only transient peaks are allowed.