Sounds that are familiar to us can show advantages in perceptual processing compared to unfamiliar sounds. This is a phenomenon indicative of perceptual learning (for review, see Wright and Zhang, 2009; Irvine, 2018; Maniglia and Seitz, 2018). The impacts of sound familiarity can appear rather quickly with initially “meaningless” sound stimuli. For instance, Agus et al. (2010) found that repetitions of random Gaussian noise samples were detected with greater sensitivity if they were the same noise samples presented on previous trials than if they were unfamiliar samples generated under the same constraints (also, see Agus and Pressnitzer, 2013). These frozen noise effects occur after a few trials (Agus et al., 2010), can be seen for noise samples as short as 10 ms (Andrillon et al., 2015), and are observable a month or more after initial exposure (Viswanathan et al., 2016; Bianco et al., 2020; Agus and Pressnitzer, 2021). Also notable is that they can occur with randomly generated tone pattern stimuli that closely resemble the acoustic characteristics of many real-world sounds and receptive field properties of auditory cortical neurons (DeCharms et al., 1998; Sohoglu and Chait, 2016; Bianco et al., 2020; Herrmann et al., 2021). This introduces exciting possibilities for studies of perceptual learning with complex sounds that are uncorrupted by previous knowledge on the part of listeners (as is the case with speech or environmental sounds).

In these previous works, sound repetitions mostly occurred consecutively with no intervening stimuli. In real-world environments, however, repetitions are rarely experienced without distracters. Examples include environmental noise during alarm sound repetition (Edworthy et al., 2018) and accompaniment to the melody of a single musical instrument (Waggoner, 2011). It is well known that detection can be hindered in these types of scenarios (Durlach, 2006). In one recent study, it was found that listeners could learn repeated tone patterns interleaved with other random patterns over the course of trials, but this learning was limited compared to patterns presented without distracters (Bianco et al., 2020). Also, this learning occurred within the repetition task itself. This leaves ambiguity as to whether learning entails memory for a specific sound pattern, or a learned strategy to listen for the sound quality that results from pattern repeats (e.g., “wooshing”; Warren and Bashford, 1981). Distinguishing these will be informative for development of perceptual learning models where these possibilities can be associated with different mechanisms (e.g., plasticity in signal representations or changes in top-down selective attention). Whether or not perceptual learning transfers to a listening situation where repeats are interleaved with similar sounds is also undetermined. Such work is needed to assess the predictions of several learning theories that the benefits of perceptual learning lie in the potential for further learning on untrained tasks, not just performance in the trained task (e.g., Gibson, 1969; Mercado III, 2008; Goldstone et al., 2010; Seppänen et al., 2013). This is especially needed in tasks that induce learning with stimuli that do not have preexisting biases associated with them (e.g., speech). The current experiment addresses both of these questions.

In a first familiarization phase, synchronous tone train patterns having randomly generated frequencies between 300 and 1,200 Hz were presented to listeners. Instructions were to answer whether a sound was “Sound A” or “Sound B.” While Sound A was frozen throughout this phase (“Familiar-Fixed” – FF), Sound B was generated randomly on each trial (“Unfamiliar-Unfixed” – UU). In a following repetition detection phase of the experiment, listeners were tasked to detect repeating patterns of synchronous tone trains within a relatively long excerpt (4.2 s in length) containing multiple tonal patterns. Previously familiarized (FF) and unfamiliar patterns (Unfamiliar – Fixed – UF; UU) were presented within excerpts. On “repeating” trials, repeating patterns were always interleaved with other randomly generated tonal patterns. It was hypothesized that familiarization with a “meaningless” sound pattern presented by itself would lead to differences in repetition detection sensitivity, response time, and learning rate between familiar and unfamiliar sounds. It was expected that this effect would be observable when patterns were interleaved with other patterns having similar characteristics.

Listeners learned the categorization task adequately in the familiarization phase. Mean sensitivity (d') to Sound A vs. Sound B differences was 1.78 (SD = 1.11), 2.82 (SD = 0.87), and 3.14 (SD = 0.75) in blocks 1–3, respectively. A sign test showed sensitivity in the last block of the familiarization phase to be significantly above zero, χ² = 30, p < 0.001. Thus, listeners moved on to the repetition detection phase with familiarity for their unique FF sound pattern.

Figure 2 shows d′ and hit RTs over the course of the repetition detection phase. For d′, there was a significant effect of window, χ² = 23.43, p < 0.001, demonstrating increased sensitivity over the course of the repetition detection task. There was also a significant sound type x window interaction, χ² = 43.58, p < 0.001, demonstrating that the rate of learning was not equal across sound types. Indeed, the model estimated learning slope for FF sounds was significantly steeper than UF, β = −0.04, t = 4.36, p < 0.001, and UU sounds, β = −0.06, t = 6.57, p < 0.001. UF sounds showed no significant difference in learning slope compared to UU sounds, p > 0.10. The effect of condition was not significant, p > 0.10.

The relatively long length of RTs shown in Figure 2 likely reflects the fact that the task was difficult and participants were instructed to value accuracy over response speed. Nevertheless, RTs on hit trials showed a significant effect of window, χ² = 16.42, p < 0.001, owing to RTs reducing in latency over the course of the repetition detection task. There was also a significant sound type x window interaction, χ² = 27.98, p < 0.001, demonstrating differences in the slope of these RT reductions. Model estimated learning slope was significantly steeper for FF sounds, β = 0.02, t = 5.22, p < 0.001, and UU sounds, β = 0.01, t = 3.98, p < 0.001, compared to UF sounds. The slope for FF and UU sounds was not significantly different, p = 0.094. The effect of condition was not significant, p > 0.10.

The current study was designed to determine whether familiarization with a meaningless sound pattern would facilitate the ability to detect repeated presentations of that pattern among acoustically similar distracters. Familiarity was induced through a categorization task in which a randomly generated pattern of multitone complexes was assigned to one category (FF type) while other randomly generated patterns (UU types) were assigned to another. Though initial sensitivity to FF sounds in a following repetition detection task was comparable to unfamiliar sounds, FF sounds showed a learning advantage. Sensitivity (d′) improved at a faster rate and reached a higher level compared to sounds that were repeated throughout the repetition detection task (UF), or that were generated randomly on each trial (UU). Response times decreased over the course of the repetition detection task, but did so at a faster rate for FF and UU sounds compared to UF sounds.

The processes leading to the familiarity effects observed here are not likely to be based in any procedural type of learning. This is because any procedural learning (e.g., developing a concept for what “repeat” means, learning the repetition rate of repeats, etc.) should have been equal across all sound types. It is also unlikely that differences among sound types were related to learning a specific repetition sound quality during the familiarization phase. This is because no consecutive presentations of the FF sound occurred at a rate as fast as that experienced in the repetition detection task (e.g., from trial-to-trial), nor were presentations isochronous during categorization training. The impacts of familiarity are likely perceptual and based in memory for the FF sound itself. This is in line with previous conclusions on learning in UF vs. UU sound type comparisons with Gaussian noise samples (Agus and Pressnitzer, 2013). Though in that study all learning did take place within the repetition detection task, those authors showed an increased false alarm rate to trials in which a single familiar pattern was presented. This showed that UF patterns were detectable even without repeats. Together, theses studies suggest that learning in repetition detection tasks reflects the learning of a specific sound pattern.

An advantage of UF over UU sound types has been consistently demonstrated in repetition detection tasks similar to the one used here (Agus et al., 2010; Agus and Pressnitzer, 2013; Andrillon et al., 2015; Viswanathan et al., 2016; Herrmann et al., 2021). However, no such advantage was observable in this study. Rather, the only difference between UF and UU sounds appeared as a disadvantage for the former in learning as measured by RT. This failure to replicate might be related to the interleaving with distracter sounds (i.e., more informational masking), the lack of immediate feedback, and/or the accompanying FT sound type. Further speculation would require an alternative experimental design. Nevertheless, the FF vs. UF comparisons shed light on the impact that familiarity with meaningless sound patterns can have on repetition detection performance and learning. That differences between FF and UF sounds appeared over the course of learning is consistent with a recurring idea in perceptual learning theory that the benefits of familiarization with a stimulus lie in the impact on learning potential (e.g., accelerated learning rates or capacity to learn; Gibson, 1969; Mercado III, 2008; Goldstone et al., 2010). Prior research on this hypothesis has mostly employed familiar speech sounds (cf. Wisniewski et al., 2014), known environmental objects (for review, see Fine and Jacobs, 2002), or preexisting individual differences in perceptual acuity (e.g., Orduña et al., 2005). Familiarity with these methods comes from an unknown learning history with learned perceptual and social biases (e.g., towards one spoken accent or another; Gluszek and Dovidio, 2010). In contrast, this study produced familiarity effects which are unlikely to have any contribution from preexisting biases of listeners. FF versus UF comparisons in tasks like the one use here will be of use to further the study of familiarity effects in perceptual learning without these confounding biases.

Future research studies are needed to tease out the roles of specific learning processes in the effect of familiarity on learning rates observed here. One possibility is that a filter is developed during the familiarization phase based on predictable sound patterns (Heilbron and Chait, 2018). This could then transfer over to the repetition detection task (e.g., Amitay et al., 2014). A filter such as this could operate to separate previously experienced patterns (signals) from novel patterns (noise). That filter could be accomplished through long-term plasticity in the auditory processing hierarchy (e.g., Wisniewski et al., 2017; Irvine, 2018) or be dependent on transient shifts in attention (Carlin and Elhilali, 2015). Another possibility is that representations of initially “meaningless” sounds build to meaning over the course of the familiarization phase. For instance, an FF sound could end up being represented by the Sound A label/category. When performing the repetition detection task, listeners may be able to use this representation at a higher-level of the auditory hierarchy to enhance learning (Ahissar et al., 2009). This study has laid out a paradigm in which these possibilities can be further investigated.

The raw data supporting the conclusions of this article will be made available by the author, without undue reservation.

The studies involving human participants were reviewed and approved by Kansas State University Institutional Review Board. The patients/participants provided their written informed consent to participate in this study.

The author confirms being the sole contributor of this work and has approved it for publication.

This research was supported by the Cognitive and Neurobiological Approaches to Plasticity (CNAP) Center of Biomedical Research Excellence (COBRE) of the National Institutes of Health under Grant No. P20GM113109. Publication of this article was funded in part by the Kansas State University Open Access Publishing Fund.

The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.