Previous work has shown that speech syllables in continuous noise are easier to recognise than the same syllables in noise which is turned on at the beginning of the syllable and off at the end (co-gated noise). A possible explanation of this is that the firing thresholds of neurones in the cochlear nerve are adapted by the continuous noise so that their dynamic ranges are more appropriate for the speech in noise ([1], [2]). If this is so noise of different durations preceding the syllables should adapt these thresholds by different amounts and so affect the recognition scores of the syllables. In order to test this hypothesis experiments have been performed to measure the recognition rates of plosive-vowel syllables as a function of preceding noise duration. It was found that noise of relatively short durations (of the order of 100 ms) decreases the recognition rate of both the plosives and the vowels but noise of longer duration increases the recognition rate. In order to account for these findings it is suggested that perceptual streaming takes place as well as noise adaptation.