Temporally Segmented Speech

Alternated and interrupted speech both show a minimum of intelligibility when the bursts of speech reaching the listener last about 150 msec. A similar effect occurs with “temporally segmented” speech, which is made by splicing silent intervals into a recording of continuous speech, thus dividing it up into “speech intervals” separated from each other by “silent intervals.” The advantage of this material is that the durations of speech and silent intervals can be independently varied, without discarding any of the speech. An earlier experiment [see MIT Res. Lab. Electron., Quart. Progr. Rep., No. 103 (1971), p. 126.] suggested that the dip in intelligibility is actually due to two separate effects: (1) the probability that a speech interval will be recognized decreases as its duration decreases and (2) the probability that signal parameters can be followed from one speech interval to the next decreases as the duration of the intervening silent interval increases. Experimental support for this view has been obtained using temporally segmented speech (1) by varying the duration of speech intervals with silent intervals held constant and (2) by varying the duration of silient intervals, with speech intervals held constant. [This research was supported by NIH Grant No. NSO4332.]