Acoustic echo cancellation for hands-free telephony using neural networks

One of the limitations of linear adaptive echo cancellers in hands-free environments is their inability to effectively cancel nonlinearities which are generated mainly in the loudspeaker during large signal peaks. The soft-clipping effect encountered when large signals are applied to the loudspeaker is modelled in a neural network using a piecewise linear/sigmoid activation function. A three-layer fully adaptive feedforward network is used to model the room/speakerphone transfer function using the special activation function. This network structure improves the ERLE performance by 10 dB at low to medium loudspeaker volumes compared to a NLMS echo canceller