Perceptual learning of noise vocoded words: Effects of feedback and lexicality.

Abstract

Speech comprehension is resistant to acoustic distortion in the input, reflecting listeners' ability to adjust perceptual processes to match the speech input. This adjustment is reflected in improved comprehension of distorted speech with experience. For noise vocoding, a manipulation that removes spectral detail from speech, listeners' word report showed a significantly greater improvement over trials for listeners that heard clear speech presentations before rather than after hearing distorted speech (clear-then-distorted compared with distorted-then-clear feedback, in Experiment 1). This perceptual learning generalized to untrained words suggesting a sublexical locus for learning and was equivalent for word and nonword training stimuli (Experiment 2). These findings point to the crucial involvement of phonological short-term memory and top-down processes in the perceptual learning of noise-vocoded speech. Similar processes may facilitate comprehension of speech in an unfamiliar accent or following cochlear implantation.

Perceptual learning of degraded speech by minimizing prediction error

“…Previous behavioral studies have shown that the provision of relevant prior knowledge enhances both immediate perceptual clarity (1,29,35,45,46) and perceptual learning of degraded speech (6,14,(16)(17)(18). However, our work goes beyond these behavioral studies by identifying a common neural signal (reductions in the STG response) that is associated with both these effects.…”

“…As described in SI Discussion, this account provides a neural implementation of "attentional weighting" theories of perceptual learning (5); with learning arising from Hebbian weight updates that minimize prediction errors when degraded speech matches prior predictions. This account therefore explains why perceptual learning of vocoded speech is enhanced when the content of degraded speech is predicted accurately (6,18); these trials lead to learning by allowing listeners to attend more appropriately to informative sensory features in degraded speech (5,7,62,63).…”

“…During the training phase, participants completed a modified version of the clarity-rating task previously used in behavioral and MEG studies combined with a manipulation of prior knowledge previously shown to enhance perceptual learning of degraded (vocoded) speech (Fig. 1B) (6,18,23). Speech was presented with three, six, or 12 channels of sensory detail.…”

“…Dual-mechanism accounts therefore propose that the influence of prior knowledge resides at a hierarchically late (e.g., decision) stage of processing and attribute the effect of learning to offline synaptic changes in earlier-level sensory cortex that take place after sensory stimulation (14,15). However, other work has shown that perceptual learning of degraded stimuli is enhanced if accurate prior knowledge is provided before the presentation of degraded or otherwise ambiguous sensory input (6,14,(16)(17)(18). Consistent with this behavioral association, alternative single-mechanism accounts have proposed that a single system containing multiple interacting levels of representation supports the effects of both prior knowledge and perceptual learning (15,19,20).…”

Perceptual learning for speech

“…However, a recent follow-up study modified this conclusion (Hervais-Adelman, Davis, Johnsrude, & Carlyon, 2008). Hervais-Adelman et al hypothesized that listeners in their previous study might have failed to learn from sentences made up of nonwords because such stimuli could be quite difficult to keep in memory, preventing perceptual learning processes from operating on them.…”