On‐line measures of comprehension of natural and synthetic speech

Abstract

Accumulated perceptual research comparing natural and synthetic speech indicates relatively large differences in tasks assessing acoustic‐phonetic processing, and small differences in tasks assessing higher levels of processing related to comprehension. Studies comparing comprehension of passages of fluent natural and synthetic speech have generally examined performance on questions presented after subjects have listened to a passage. Such postperceptual measures are known to be relatively insensitive to differences in “real‐time” processing operations. The present investigation employed an “on‐line” measure of processing, i.e., word monitoring, to study comprehension. Subjects in these studies were presented with three types of passages—(1) natural speech, (2) high‐quality synthetic speech, and (3) low‐quality synthetic speech—and were required to monitor for target words as well as verify postperceptual comprehension questions. Monitoring latencies and verification performance will be discussed in terms of differences in perceptual processing and comprehension of natural and synthetic speech. [Work supported by NSF.]