Comparison of acoustic features of time-compressed and natural speech

Abstract

Time-compressed speech is used as a speech material to examine the upper limits of processing speed in young and old listeners with and without hearing loss. Generally, older normal-hearing listeners at 60% time compression perform similarly to younger normal-hearing listeners at 70% time compression on speech recognition test. This may be due to slowing of processing speed that typically accompanies aging. It has not been determined if time-compressed speech material carries spectral distortions or other acoustic changes that affect the ability to process rapidly. If so, differences in performance between younger and older listeners may be attributable in part to acoustic distortions other than temporal changes. In this work, we examine acoustic feature changes in time-compressed speech processed using synchronized overlap-add (SOLA) method. Also, we discuss which parts of speech (consonants or vowels) are more affected by time compression. In the temporal domain, we compare the duration of consonants, vowels and voice onset times (VOT) in both 50% compressed speech and natural speech. We also examine the long-term power spectra of both signals. Possible effects on speech intelligibility will be discussed.