Non-linear processing in cochlear spaced sub-bands using artificial neural networks for multi-microphone adaptive speech enhancement

A general class of single hidden-layered, linear-in-the-parameters feedforward Artificial Neural Networks is proposed for processing band-limited signals in a multi-microphone sub-band adaptive speech enhancement scheme. The sub-band spacing within the adaptive speech enhancement system is set according to a published cochlear function. Comparative results achieved in simulation experiments demonstrate that the proposed sub-band scheme is capable of significantly outperforming conventional linear processing based wide-band and sub-band noise cancellation methods, in the presence of non-linear interference.