It shouldn’t necessarily, you can find out how much of the speaker output ends up in the microphone input and compensate. I’m sure the models are more complicated to deal with echoes and distortion but the same approach should basically work: fit the filter offline, apply it online.