This paper describes an implementation of a speech recognition system using a neural network, originally proposed by Kurogi [1], which is based on findings from the auditory mechanism, including both auditory signal processing and afferent pathway signal processing. A design procedure for a multi-rate cochlear model is described. The model uses a cascade of constant-Q filters. The model covers the bandwidth from 75Hz to 3.55kHz and is used as a pre-processor to the neural network. The neural network reproduces response patterns typical in the auditory system. The network does not use any time-warping and makes no use of the back-propagation algorithm for training. Preliminary results are presented for speaker independent speech recognition experiments carried out using a subset of the British Telecom Sl-Connex database.