One way towards an understanding of speech perception is by means of computational models based on neuroanatomical and neurophysiological data. These can be used to study the processing of speech sounds in the auditory system and the results can be compared with data on human perception. A model of the cochlear nerve and cochlear nucleus is described. It is suggested that some units in the cochlear nucleus, choppers, selectively extract a good frequency representation of the incoming signals whereas others, onset units, form a good time representation. The output of the onset units has been used to estimate the pitch of speech sounds. An extension of the model has been used to generate amplitude modulation maps similar to those found in the inferior colliculus. This has been employed in a vowel segregation experiment and gave similar results to those obtained by human perception of the same stimuli.