Tales from the dip: Factors in cross-rate intelligibility variation of interrupted speech

Abstract

Speech intelligibility involves integration of temporally and spectrally distributed acoustic information into higher-order perceptual categories to obtain individual words. In 1950, Miller and Licklider pioneered a simple but powerful method of interrupting speech that has been extensively used to investigate factors that make speech signals perceptually robust. Among numerous subsequent studies, a consistent finding has been a nonmonotonic relationship between intelligibility and interruption rate. As interruption rate increases, and the duration of speech fragments in each interruption cycle decreases, a U-shaped rate-intelligibility function with a dip around 1–5 Hz frequently emerges. While many factors (e.g., speech materials, task parameters, listener age, or hearing status) have been shown to influence performance at specific interruption rates, reasons for the appearance and location of the dip in the function have remained obscure. Previous work indicates that the location of the dip in the rate-intelligibility function can be altered predictably with changes in the temporal structure of the interrupted speech stream, and may vary with the duration of the corresponding perceptual units. These findings will be considered in the context of neurophysiological and information-processing models of interrupted speech, and used to suggest a framework to guide future research and practical applications.