Perceptual learning of degraded speech by minimizing prediction error

Abstract

Human perception is shaped by past experience on multiple timescales. Sudden and dramatic changes in perception occur when prior knowledge or expectations match stimulus content. These immediate effects contrast with the longer-term, more gradual improvements that are characteristic of perceptual learning. Despite extensive investigation of these two experience-dependent phenomena, there is considerable debate about whether they result from common or dissociable neural mechanisms. Here we test single-and dualmechanism accounts of experience-dependent changes in perception using concurrent magnetoencephalographic and EEG recordings of neural responses evoked by degraded speech. When speech clarity was enhanced by prior knowledge obtained from matching text, we observed reduced neural activity in a peri-auditory region of the superior temporal gyrus (STG). Critically, longer-term improvements in the accuracy of speech recognition following perceptual learning resulted in reduced activity in a nearly identical STG region. Moreover, short-term neural changes caused by prior knowledge and longer-term neural changes arising from perceptual learning were correlated across subjects with the magnitude of learninginduced changes in recognition accuracy. These experience-dependent effects on neural processing could be dissociated from the neural effect of hearing physically clearer speech, which similarly enhanced perception but increased rather than decreased STG responses. Hence, the observed neural effects of prior knowledge and perceptual learning cannot be attributed to epiphenomenal changes in listening effort that accompany enhanced perception. Instead, our results support a predictive coding account of speech perception; computational simulations show how a single mechanism, minimization of prediction error, can drive immediate perceptual effects of prior knowledge and longer-term perceptual learning of degraded speech.perceptual learning | predictive coding | speech perception | magnetoencephalography | vocoded speech S uccessful perception in a dynamic and noisy environment critically depends on the brain's capacity to change how sensory input is processed based on past experience. Consider the way in which perception is enhanced by accurate prior knowledge or expectations. Sudden and dramatic changes in subjective experience can occur when a distorted and otherwise unrecognizable perceptual object is seen or heard after the object's identity is revealed (1-4). Such effects occur almost immediately; striking changes in perceptual outcomes occur over a timescale of seconds or less. However, not all effects of past experience emerge as rapidly as these effects of prior knowledge. With perceptual learning, practice in perceiving certain types of stimuli results in gradual and incremental improvements in perception that develop over a timescale of minutes or longer (Fig. 1A) (5, 6, 7). Critically, perceptual learning can generalize beyond the stimuli experienced during training, e.g., to visual forms presented in dif...

Donepezil enhances understanding of degraded speech in Alzheimer's disease

“…These findings are in line with other work demonstrating cholinergic effects on visual perception and cortical function in AD,8 with retained endogenous perceptual learning mechanisms 15. While the mechanism of the procholinergic effect on degraded speech perception remains undefined, there are several plausible candidates, acting alone or in concert: these include enhanced precision of synaptic transmission and predictive filtering in ascending auditory pathways and facilitation of spectral integration, feature encoding and tracking in auditory cortex 6, 16, 17, 18, 19, 20…”

Pure linguistic interference during comprehension of competing speech signals

“…The intelligibility of the NV sentences was low before training [mean ¼ 26.07%, standard error (SE) ¼ 1.91] and improved after training (mean-¼ 56.34%, SE ¼ 2.59). The intelligibility of untrained 4-band NV sentences also improved (relative to the matched pre-training sentences) after training (mean-¼ 40.50%, SE ¼ 2.10), which suggests that the training could generalize to new speech utterances, in line with previous reports (Davis et al, 2005;Sohoglu and Davis, 2016). Both improvements were significant [pre-vs post-trained: t(23) ¼ 18.34, p < 0.001; prevs post-untrained: t(23) ¼ 10.14, p < 0.001, Bonferroni corrected], but intelligibility after training was stronger for the trained sentences compared to the untrained sentences [t(23) ¼ 10.64, p < 0.001].…”

“…Crucially, the NV speech sentences were initially poorly intelligible but could be understood after training (Davis et al, 2005;Sohoglu and Davis, 2016). Hence, before and after training, the NV speech would have the same acoustic information but not carry the same linguistic information.…”

The Goldilocks Zone of Perceptual Learning

“…Kraljic et al, 2008b). These findings are reminiscent of Sohoglu and Davis (2016), who found that listeners learned best from partially intelligible (6-channel speech) than from unintelligible (1-channel) or highly intelligible (24-channel) speech. It should be noted that learning of complete remappings has been found in several studies (Sumner, 2011;Weatherholtz, 2015) for different contrasts (bilabial stops; vowels) in different paradigms (exposure to the deviant pronunciations via accent rating; narrative exposure), indicating that future work should tease apart variables responsible for differences across studies.…”

“…Our goal here is to assess how the typicality of the pronunciation variant affects perceptual learning. There is already empirical evidence that phonetic variation that deviates considerably from a canonical pronunciation (i.e., with heavily accented items in a nonnative accent, Witteman et al, 2013), that interferes with lexical recognition (i.e., when listeners have lower thresholds in a lexical decision task, Scharenborg & Janse, 2013), or that is less intelligible because of signal degradation (i.e., one-channel speech, Sohoglu & Davis, 2016) does not lead to perceptual learning. Further, Ganong's (1980) finding that listeners' boundaries shift more in accordance with lexical bias at their own perceptual category boundary than at continuum end points suggests different amounts of retuning across degrees of ambiguity.…”