Variables that influence the recognition performance of interrupted words: rise-fall shape and temporal location of...

The abrupt transition of a signal from off to on and vice versa typically produces spectral splatter that can mask other signals that are spectrally removed from the nominal signal frequency. Both the Miller and Licklider (1950) and Cherry (1953) studies of interrupted speech and alternated speech, respectively, acknowledged the generation of extraneous noise by the rapid on and off characteristics of their unshaped signals but noted for slower interruption rates (e.g., 10 interruptions per second); the masking effects were minimal. Recent studies of interrupted speech have avoided this issue by shaping the rise-fall times with a digital algorithm (e.g., Jin and Nelson, 2010; Wang and Humes, 2010). A second variable in the interrupted speech paradigm is the temporal location or placement of the interruptions (i.e., where in the waveform the interruptions occur). Here the issue is this: what parts of an utterance are necessary to enable intelligibility (e.g., Fogerty and Kewley-Port, 2009)? Interruptions may or may not disturb these necessary cues.

Here is the prompting question: do shaped and unshaped rise-fall characteristics of the on-segments of interrupted speech produce the same or different recognition performances? A second question arises: are recognition performances on complementary halves of an interrupted signal the same or different?

This study used a mixed-model design with two within-subject variables (unshaped and shaped rise-fall characteristic, complementary halves) and one between-subjects variable (listener group).

A total of 12 young listeners (age range: 19-29 yr) with normal hearing and 12 older listeners (age range: 53-80 yr) with hearing loss for pure tones participated.

A total of 95 consonant-vowel nucleus-consonant words were interrupted (10 interruptions per second; 50% duty cycle) by parsing alternate 50 msec segments to separate files, which provided complementary temporal halves of the target word referenced to word onset; the first on-segment of the 0 msec condition started at word onset, whereas the first on-segment of the 50 msec condition started 50 msec after word onset. The interruption routine either applied no shaping of the 4 msec rise-fall times or a cos² shape. Each listener received 25 practice words then a unique randomization of 280 interrupted words (70 words, 2 rise-fall shapes, and 2 interrupt onset conditions).

The listeners with normal hearing performed 8-16% better on the various comparable conditions than did the older listeners with hearing loss. The mean performance differences between shaped and unshaped rise-fall characteristics ranged from <1-3% and were not significant. Performance was significantly 10-17% better on the 0 msec condition than on the 50 msec condition. There was no significant interaction between the two main variables, rise-fall shape, and onset time of the interruptions.

The rise-fall shape of the onset and offset of the on-segment of the interruption cycle does not affect recognition performance of words. The location of the interruptions in a word can have a significant effect on recognition performance.