The question could be : why is AudioIODeviceCombiner class needed in the first place? CoreAudio layer allows to aggregate separated devices, so this could be entirely avoided (see this thread Estimating JUCE behavior to develop a generic architecture file for Faust)