Long Term Memory for Noise: Evidence of Robust Encoding of Very Short Temporal Acoustic Patterns

Abstract

Recent research has demonstrated that humans are able to implicitly encode and retain repeating patterns in meaningless auditory noise. Our study aimed at testing the robustness of long-term implicit recognition memory for these learned patterns. Participants performed a cyclic/non-cyclic discrimination task, during which they were presented with either 1-s cyclic noises (CNs) (the two halves of the noise were identical) or 1-s plain random noises (Ns). Among CNs and Ns presented once, target CNs were implicitly presented multiple times within a block, and implicit recognition of these target CNs was tested 4 weeks later using a similar cyclic/non-cyclic discrimination task. Furthermore, robustness of implicit recognition memory was tested by presenting participants with looped (shifting the origin) and scrambled (chopping sounds into 10- and 20-ms bits before shuffling) versions of the target CNs. We found that participants had robust implicit recognition memory for learned noise patterns after 4 weeks, right from the first presentation. Additionally, this memory was remarkably resistant to acoustic transformations, such as looping and scrambling of the sounds. Finally, implicit recognition of sounds was dependent on participant's discrimination performance during learning. Our findings suggest that meaningless temporal features as short as 10 ms can be implicitly stored in long-term auditory memory. Moreover, successful encoding and storage of such fine features may vary between participants, possibly depending on individual attention and auditory discrimination abilities. Significance Statement Meaningless auditory patterns could be implicitly encoded and stored in long-term memory.Acoustic transformations of learned meaningless patterns could be implicitly recognized after 4 weeks.Implicit long-term memories can be formed for meaningless auditory features as short as 10 ms.Successful encoding and long-term implicit recognition of meaningless patterns may strongly depend on individual attention and auditory discrimination abilities.

Keywords: STDP; implicit learning; long-term memory; meaningless stimuli; temporal resolution.

Figure 1

Exemplars of 1-s Gaussian white noises (sampling frequency = 44.1 kHz) and acoustic transformations used in the experiment. (A) Cyclic noise (CN) vs. non-cyclic noise (N): Gaussian noises typically show small amplitude variations over time. The first and second halves of a CN are identical, while an N is completely random. (B) Transformations used to loop and scramble the learned CNs in the testing session. For looping, a random time point was chosen in the first half of the sound and the sound portion preceding this time point was shifted to the end. For scrambling, the first half of the cyclic sound was cut into segments of 20 ms for version 1 and 10 ms for version 2, the segments were randomly shuffled and the resulting 500-ms sound was played back to back to create a scrambled CN. (C) Looped and Scrambled sounds: amplitude variations over time of exemplar looped and scrambled (20 ms) versions of the CN shown in (A). The color scheme of (A,C) is graded as a function of sound amplitudes, in order to facilitate identification of repeating features.

Figure 2

Changes in frequency features—in low, mid and high frequency bands—of a CN due to looping and scrambling with increasing bin sizes. The maximal frequency on the X axis corresponds to the Nyquist frequency (22,050 Hz) and the spectrum amplitude difference between original and looped/scrambled versions of a CN is plotted on the Y axis. With decreasing bin size, the difference between the resulting scrambled sound and the original sound increases, leading to greater difference in amplitude spectrum from the original, across all the frequency bands.

Figure 3

Learning session results for both sets of 10 target CNs: Each target CN was learned by a variable percentage of participants, i.e., there were no target CNs that were systematically learned by all participants.

Figure 4

Discrimination performance for intact learned target CNs vs. novel CNs in the testing session. (A) Relationship between discrimination rates of learned target and novel CNs in each participant. Participants above the diagonal show higher rates for learned vs. novel CNs, suggesting that memory facilitated the discrimination task. (B) Discrimination rates of learned target and novel CNs over time (10 blocks).

Figure 5

Discrimination performance during the testing session. (A) Performance for intact, looped, scrambled learned target CNs and novel CNs (n = 25) and (B) Discrimination performance for scrambled trials with 20 and 10 ms bin sizes (n = 16 in version 1 and n = 9 in version 2).

Figure 6

Relationship between discrimination rates of CNs in the testing session and learning efficiency (represented as a') for all participants (n = 25).