Acoustic echo cancellation using deep cerebellar model articulation controller
Shih-Wei Lan, Yu Tsao, Junghsi Lee · 2017
In this paper, we propose to adopt the deep cer-ebellar model articulation controller (DCMAC) model for acoustic echo cancellation (AEC). The DCMAC model is formed by stacking multiple CMAC models. The deep structure of the DCMAC model can characterize nonlinear transformations more effectively when compared with the conventional CMAC model. Experimental results showed that DCMAC outperforms CMAC in terms of MSE values, confirming that DCMAC yields improved capability of modeling channel characteristics.