We describe the representation of synthetic stop-consonants in a computational model of the mammalian dorsal cochlear nucleus. The speech stimuli have different values of voice-onset time (VOT) and are labelled by adult listeners as either /ga/ or /ka/, with a phonetic boundary at 44 ms VOT. The responses of the model's Type IV units to these stimuli also fall into two clear categories with a boundary at 45 ms VOT. These results provide evidence that the categorical perception of voicing in initial English stops - observed in behavioural experiments using human subjects (infant and adult) and chinchillas - may arise as a consequence of the representation of these sounds in the mammalian auditory nervous system at the level of the dorsal acoustic stria.