A preliminary investigation into the suitability of Time Encoded Speech (TES) descriptors for the segmentation of speech into consistent acoustic events is reported upon. Segmentation is accomplished by a system of parallel seekers, each one optimised to detect a particular class of acoustic event. By this method A -Matrices, that is to say second order TES symbol distribution descriptors, may be compiled for each segment and utterances compared using these as a basis. Word comparisons are achieved by a simple dynamic programming algorithm applied to the small number of segments produced.