We estimate the amount of time taken by listeners to perceive artificial stop consonants to be in the vicinity of 120 msec. The duration of the formant transitions in the syllables we employ, on the other hand, is 50 msec (followed by a nondistinctive vowel). Processing thus outlasts relevant input by some 70 msec. Moreover, forward masking develops as consonant processing takes place, which could lead to masking within a single syllable and block further‐input. These results suggest that, for a significant fraction of the time, normal speech perception takes place in the absence of input. A view of speech perception compatible with such a limitation ascribes most of the perceptual structure of speech to largely autonomous internal processes, rather than to the acoustic structure of the speech signal. This signal serves primarily to trigger the process of speech perception in this view.