Nonlinear sub-band processing for binaural adaptive speech-enhancement

Nonlinear processing using a general class of feedforward, single hidden-layered artificial neural networks is proposed for processing bandlimited noisy signals in a binaural adaptive speech-enhancement scheme. The sub-band spacing within the adaptive speech-enhancement system can be set as in human hearing, according to a published cochlear function. Preliminary comparative results achieved in simulation experiments demonstrate that the proposed binaural sub-band scheme is capable of significantly outperforming conventional linear-processing based wide-band and sub-band noise-cancellation methods, in the presence of nonlinear interference.