Segmental durations in connected speech signals

From the readings, by many talkers, of two scripts of approximately 300 syllables, two groups of seven talkers were selected on the basis of overall reading time to form a group of fast talkers and a group of slow ones. The phonetic makeup of each script and of the first two sentences of each script were analyzed to determine if they represented American text. The entire readings were segmented, using computer graphics and digital playback, into breath groups and pauses; and the first two sentences of each into phones. Average numbers and durations of breath groups and pauses were computed as a basis for categorizing talkers according to their natural tempo. (These data can be compared to those based on “unnatural” tempos obtained by having individuals speed or slow their reading.) Differences between the groups were also noted in the average durations of various categories both in and out of context. Finally, probability distributions were obtained for the durations of various categories.