Acoustic echo cancellation (AEC) algorithms need to build an adaptive model of the speaker output signal and subtract the predicted value from the microphone input in real time, preventing the model from hearing itself and falling into a feedback loop.