Marwan Jabri, Raymond Wang
State-of-the-art speech processors in cochlear implants perform channel selection using a spectral maxima strategy. This strategy can lead to confusions when high frequency features are needed to discriminate between sounds. We present in this paper a novel channel selection strategy based upon pattern recognition which al(cid:173) lows "smart" channel selections to be made. The proposed strategy is implemented using multi-layer perceptrons trained on a multi(cid:173) speaker labelled speech database. The input to the network are the energy coefficients of N energy channels. The output of the system are the indices of the M selected channels. We compare the performance of our proposed system to that of spectral maxima strategy, and show that our strategy can produce significantly better results.