Auditory memory for random time patterns

Abstract

The acquisition of auditory memory for temporal patterns was investigated. The temporal patterns were random sequences of irregularly spaced clicks. Participants performed a task previously used to study auditory memory for noise [Agus, Thorpe, and Pressnitzer (2010). Neuron 66, 610-618]. The memory for temporal patterns displayed strong similarities with the memory for noise: temporal patterns were learnt rapidly, in an unsupervised manner, and could be distinguished from statistically matched patterns after learning. There was, however, a qualitative difference from the memory for noise. For temporal patterns, no memory transfer was observed after time reversals, showing that both the time intervals and their order were represented in memory. Remarkably, learning was observed over a broad range of time scales, which encompassed rhythm-like and buzz-like temporal patterns. Temporal patterns present specific challenges to the neural mechanisms of plasticity, because the information to be learnt is distributed over time. Nevertheless, the present data show that the acquisition of novel auditory memories can be as efficient for temporal patterns as for sounds containing additional spectral and spectro-temporal cues, such as noise. This suggests that the rapid formation of memory traces may be a general by-product of repeated auditory exposure.

Long-term implicit memory for sequential auditory patterns in humans

“…In contrast, here we focus on fast memory formation for sequences of discrete tones, distinguishable only by their specific order, and presented in a surrounding context of highly similar patterns (all sequences consisted of the same 20 'building blocks'). We showed that the vast majority of patterns were learned, revealing high sensitivity to reoccurring arbitrary frequency patterns despite the exceedingly rare reoccurrence rate (every ~2.7 minutes; 5% of trials; in contrast to the much more frequent reoccurrence (< ~15 seconds) in Agus et al (2010) and Kang et al (2017).…”

“…Inspections of the nature of this memory revealed that it was robust to time reversal, and even to scrambling into bins as small as 10-20 ms (Agus, Thorpe, & Pressnitzer, 2010;Viswanathan, Rémy, Bacon-Macé, & Thorpe, 2016), indicating that the remembered features reflect local spectrotemporal idiosyncrasies within the reoccurring noise snippet. The apparent dependence of this memory on certain local features of the noise signal may also explain the high inter-sample variability often seen with this paradigm (i.e., the distinction between 'memorable' and 'not memorable' patterns; Agus et al, 2010;Viswanathan et al, 2016;Kang et al, 2017).…”

“…Cowan, 1998) argued that such sensory traces are characterized by short retention times and computational encapsulation, a large body of work has since revealed that observers can retain detailed sensory information implicitly, over long periods (Arciuli & Simpson, 2012;Chun, 2000;Jiang, Song, & Rigas, 2005;Kim, Seitz, Feenstra, & Shams, 2009;Lu & Vicario, 2014;Winkler & Cowan, 2005). A compelling instance of this was demonstrated by Agus et al (2010; see also Agus & Pressnitzer, 2013;Kang, Agus, & Pressnitzer, 2017) who showed that naïve listeners readily remembered certain spectro-temporal features of random noise bursts, such that reoccurring snippets were recognized weeks after initial exposure.…”

“…Participants were naïve to the experimental manipulation. It was expected that, if implicit memory is specific to the sequential structure of regularity, the RT advantage should disappear in the time-reversed block (see also Kang et al, 2017).…”

“…Agus Kang et al (2017) proposed that incorporating a degree of temporal integration can also account for similar effects observed with random click trains. It is possible that state-dependent neural place maps, as proposed by Kang et al (2017; see also Karmarkar & Buonomano, 2007;Lim, Lagoy, Shinn-Cunningham, & Gardner, 2017) incorporating an integration time of several hundred msec, may also underpin memory for discrete tone sequences. As will be discussed further below, the behavioural pattern reported in the present paper is consistent with sequential information being stored as short sub-sequences, i.e.…”