Date Published: October 24, 2018
Publisher: Public Library of Science
Author(s): Yafit Gabay, Avi Karni, Karen Banai, Jérôme Prado.
Learning to decipher acoustically distorted speech serves as a test case for the study of language-related skill acquisition in persons with developmental dyslexia (DD). Deciphering this type of input is rarely learned explicitly and does not yield conscious insights. Problems in implicit and procedural skill learning have been proposed as possible causes of DD. Here we examined the learning of time-compressed (accelerated) speech and its generalization to novel materials among young adults with DD compared to typical readers (TD). All participants completed a training session that involved judging the semantic plausibility of sentences, during which the level of time-compression was changed using an adaptive (staircase) procedure according to each participant’s performance. In the test, phase learning (test on same items) and generalization (test on new items and same items spoken by a new speaker) were assessed. Both groups showed robust gains after training. Moreover, after training, the initial disadvantage of the DD group was no longer significant. After training, both groups experienced relative difficulties in deciphering learned tokens spoken by a different voice, though participants with DD were less able to generalize the gains to deciphering new tokens. Thus, DD individuals benefited from repeated experience with time-compressed speech no less than typical readers, but their evolving skill was apparently more dependent on the specific characteristics of the tokens. Atypical generalization, which indicates that perceptual learning is contingent on lower-level features of the input though does not necessarily point to impaired learning potential per se, may explain some of the contradictory findings in published studies of speech perception in DD.
Although usually transparent to listeners, speech perception is quite a challenging task. In particular, it requires mapping the acoustic input onto stable (pre-lexical/lexical) representations even though the speech signal itself is variable as a result of between-speaker differences, changes in speech rate  and environmental conditions . Speech stimuli constitute a learning challenge for the perceptual system because accurate speech recognition requires generalization across the highly variable acoustic information that underlies the speech signal. Listeners are capable of overcoming these variations in speech through perceptual learning, according to which they align their perceptual system with new variations in the speech input . Perceptual learning has been demonstrated across a variety of tasks in which the speech signal is noisy, distorted (e.g., noise vocoded, spectrally shifted or time-compressed speech) or otherwise unusual (e.g., unfamiliar dialects or accents) . Previous studies suggest that adaptive training procedures that start off with relatively little signal distortion (“easy” items, not far removed from standard speech) may be advantageous for learning and its generalization [5, 6].
Impairments in implicit skill acquisition have been proposed to have a deleterious impact in DD [13, 18–21, 24, 62]. Perceptual learning of speech represents a case of procedural learning (i.e., skill learning—how to, what to do knowledge—that are acquired implicitly and are difficult to verbalize explicitly, . The current results show that despite the initial advantage of typical readers over struggling readers in the ability to decipher time-compressed speech, both groups improved with practice, such that the magnitude of learning was similar in the two groups. Thus, given an identical training experience in deciphering time-compressed speech, young adults with DD were as adept in acquiring the specific skill as their typical reading peers. Moreover, compared to their pre-training baseline performance, both groups improved in deciphering tokens uttered in a new (untrained) speaker’s voice and much improved in their ability to decipher new time-compressed tokens. Nevertheless, listeners with DD were less able to transfer their learning-related gains to tokens that were not encountered during the training session. Both groups were hampered in deciphering the trained tokens delivered by a new speaker compared to their ability to decipher tokens presented in the familiar (trained) voice.