A common element among several proposed speech perception models is that an acoustic representation of the speech signal persists briefly in memory prior to phonetic processing. It is also commonly hypothesized that with the passage of time the auditory signals are degraded. These hypotheses were tested in a perceptual experiment in which subjects were presented short segments of vowels for identification. A single glottal pulse from each of the vowels /ɑ, i, I, æ, ε,/ was excised from natural speech and repeated to produce three types of stimuli: two contiguous glottal pulses, four contiguous glottal pulses, and two contiguous glottal pulses followed by two more contiguous glottal pulses with interspersed silences of 10, 20, 40, 60, 80, 120, 240, 480, 1000, and 2000 ms. The stimuli for all vowels and types were randomized and presented to subjects. As expected, the four contiguous glottal pulse stimuli gave better recognition scores than the two contiguous glottal pulse stimuli. It was predicted that as the interstimulus silent intervals increased for the third class of stimuli that the recognition scores would approach those of the two glottal pulse stimuli. In general, this hypothesis was confirmed except for an apparent perceptual recovery interval between 60 and 240 ms.