Implementation of a masking network for speech perception

In this paper, we propose an asymmetric winner-take-all (WTA) network and its VLSI implementation. This circuit implements the masking effect in human speech perception. Simulation results on real speech data shows that this network not only performs masking effect but also extracts spectral envelope and formant structure, in speech signals even under severe noisy environments. Finally, a 256-node VLSI implementation of the proposed network is also presented.